diff options
Diffstat (limited to 'usr/src/uts/common/syscall')
| -rw-r--r-- | usr/src/uts/common/syscall/brandsys.c | 8 | ||||
| -rw-r--r-- | usr/src/uts/common/syscall/fcntl.c | 137 | ||||
| -rw-r--r-- | usr/src/uts/common/syscall/memcntl.c | 11 | ||||
| -rw-r--r-- | usr/src/uts/common/syscall/open.c | 8 | ||||
| -rw-r--r-- | usr/src/uts/common/syscall/rusagesys.c | 14 | ||||
| -rw-r--r-- | usr/src/uts/common/syscall/rw.c | 222 | ||||
| -rw-r--r-- | usr/src/uts/common/syscall/sendfile.c | 19 | ||||
| -rw-r--r-- | usr/src/uts/common/syscall/sysconfig.c | 24 | ||||
| -rw-r--r-- | usr/src/uts/common/syscall/uadmin.c | 6 |
9 files changed, 365 insertions, 84 deletions
diff --git a/usr/src/uts/common/syscall/brandsys.c b/usr/src/uts/common/syscall/brandsys.c index 9b4bd38baa..8ee5511fd0 100644 --- a/usr/src/uts/common/syscall/brandsys.c +++ b/usr/src/uts/common/syscall/brandsys.c @@ -23,7 +23,9 @@ * Use is subject to license terms. */ -#pragma ident "%Z%%M% %I% %E% SMI" +/* + * Copyright (c) 2014, Joyent, Inc. All rights reserved. + */ #include <sys/brand.h> #include <sys/systm.h> @@ -35,7 +37,7 @@ */ int64_t brandsys(int cmd, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3, - uintptr_t arg4, uintptr_t arg5, uintptr_t arg6) + uintptr_t arg4, uintptr_t arg5) { struct proc *p = curthread->t_procp; int64_t rval = 0; @@ -49,7 +51,7 @@ brandsys(int cmd, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3, return (set_errno(ENOSYS)); if ((err = ZBROP(p->p_zone)->b_brandsys(cmd, &rval, arg1, arg2, arg3, - arg4, arg5, arg6)) != 0) + arg4, arg5)) != 0) return (set_errno(err)); return (rval); diff --git a/usr/src/uts/common/syscall/fcntl.c b/usr/src/uts/common/syscall/fcntl.c index 7421957235..d631fe62f6 100644 --- a/usr/src/uts/common/syscall/fcntl.c +++ b/usr/src/uts/common/syscall/fcntl.c @@ -22,6 +22,7 @@ /* * Copyright (c) 1994, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. + * Copyright 2015, Joyent, Inc. */ /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */ @@ -53,7 +54,8 @@ #include <sys/cmn_err.h> -static int flock_check(vnode_t *, flock64_t *, offset_t, offset_t); +/* This is global so that it can be used by brand emulation. */ +int flock_check(vnode_t *, flock64_t *, offset_t, offset_t); static int flock_get_start(vnode_t *, flock64_t *, offset_t, u_offset_t *); static void fd_too_big(proc_t *); @@ -271,11 +273,12 @@ fcntl(int fdes, int cmd, intptr_t arg) * The file system and vnode layers understand and implement * locking with flock64 structures. So here once we pass through * the test for compatibility as defined by LFS API, (for F_SETLK, - * F_SETLKW, F_GETLK, F_GETLKW, F_FREESP) we transform - * the flock structure to a flock64 structure and send it to the - * lower layers. Similarly in case of GETLK the returned flock64 - * structure is transformed to a flock structure if everything fits - * in nicely, otherwise we return EOVERFLOW. + * F_SETLKW, F_GETLK, F_GETLKW, F_OFD_GETLK, F_OFD_SETLK, F_OFD_SETLKW, + * F_FREESP) we transform the flock structure to a flock64 structure + * and send it to the lower layers. Similarly in case of GETLK and + * OFD_GETLK the returned flock64 structure is transformed to a flock + * structure if everything fits in nicely, otherwise we return + * EOVERFLOW. */ case F_GETLK: @@ -283,6 +286,11 @@ fcntl(int fdes, int cmd, intptr_t arg) case F_SETLK: case F_SETLKW: case F_SETLK_NBMAND: + case F_OFD_GETLK: + case F_OFD_SETLK: + case F_OFD_SETLKW: + case F_FLOCK: + case F_FLOCKW: /* * Copy in input fields only. @@ -345,20 +353,65 @@ fcntl(int fdes, int cmd, intptr_t arg) if ((error = flock_check(vp, &bf, offset, maxoffset)) != 0) break; + if (cmd == F_FLOCK || cmd == F_FLOCKW) { + /* FLOCK* locking is always over the entire file. */ + if (bf.l_whence != 0 || bf.l_start != 0 || + bf.l_len != 0) { + error = EINVAL; + break; + } + if (bf.l_type < F_RDLCK || bf.l_type > F_UNLCK) { + error = EINVAL; + break; + } + } + + if (cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) { + /* + * TBD OFD-style locking is currently limited to + * covering the entire file. + */ + if (bf.l_whence != 0 || bf.l_start != 0 || + bf.l_len != 0) { + error = EINVAL; + break; + } + } + /* * Not all of the filesystems understand F_O_GETLK, and * there's no need for them to know. Map it to F_GETLK. + * + * The *_frlock functions in the various file systems basically + * do some validation and then funnel everything through the + * fs_frlock function. For OFD-style locks fs_frlock will do + * nothing so that once control returns here we can call the + * ofdlock function with the correct fp. For OFD-style locks + * the unsupported remote file systems, such as NFS, detect and + * reject the OFD-style cmd argument. */ if ((error = VOP_FRLOCK(vp, (cmd == F_O_GETLK) ? F_GETLK : cmd, &bf, flag, offset, NULL, fp->f_cred, NULL)) != 0) break; + if (cmd == F_FLOCK || cmd == F_FLOCKW || cmd == F_OFD_GETLK || + cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) { + /* + * This is an OFD-style lock so we need to handle it + * here. Because OFD-style locks are associated with + * the file_t we didn't have enough info down the + * VOP_FRLOCK path immediately above. + */ + if ((error = ofdlock(fp, cmd, &bf, flag, offset)) != 0) + break; + } + /* * If command is GETLK and no lock is found, only * the type field is changed. */ - if ((cmd == F_O_GETLK || cmd == F_GETLK) && - bf.l_type == F_UNLCK) { + if ((cmd == F_O_GETLK || cmd == F_GETLK || + cmd == F_OFD_GETLK) && bf.l_type == F_UNLCK) { /* l_type always first entry, always a short */ if (copyout(&bf.l_type, &((struct flock *)arg)->l_type, sizeof (bf.l_type))) @@ -387,7 +440,7 @@ fcntl(int fdes, int cmd, intptr_t arg) obf.l_pid = (int16_t)bf.l_pid; if (copyout(&obf, (void *)arg, sizeof (obf))) error = EFAULT; - } else if (cmd == F_GETLK) { + } else if (cmd == F_GETLK || cmd == F_OFD_GETLK) { /* * Copy out SVR4 flock. */ @@ -591,6 +644,11 @@ fcntl(int fdes, int cmd, intptr_t arg) case F_SETLK64: case F_SETLKW64: case F_SETLK64_NBMAND: + case F_OFD_GETLK64: + case F_OFD_SETLK64: + case F_OFD_SETLKW64: + case F_FLOCK64: + case F_FLOCKW64: /* * Large Files: Here we set cmd as *LK and send it to * lower layers. *LK64 is only for the user land. @@ -611,6 +669,16 @@ fcntl(int fdes, int cmd, intptr_t arg) cmd = F_SETLKW; else if (cmd == F_SETLK64_NBMAND) cmd = F_SETLK_NBMAND; + else if (cmd == F_OFD_GETLK64) + cmd = F_OFD_GETLK; + else if (cmd == F_OFD_SETLK64) + cmd = F_OFD_SETLK; + else if (cmd == F_OFD_SETLKW64) + cmd = F_OFD_SETLKW; + else if (cmd == F_FLOCK64) + cmd = F_FLOCK; + else if (cmd == F_FLOCKW64) + cmd = F_FLOCKW; /* * Note that the size of flock64 is different in the ILP32 @@ -636,18 +704,65 @@ fcntl(int fdes, int cmd, intptr_t arg) if ((error = flock_check(vp, &bf, offset, MAXOFFSET_T)) != 0) break; + if (cmd == F_FLOCK || cmd == F_FLOCKW) { + /* FLOCK* locking is always over the entire file. */ + if (bf.l_whence != 0 || bf.l_start != 0 || + bf.l_len != 0) { + error = EINVAL; + break; + } + if (bf.l_type < F_RDLCK || bf.l_type > F_UNLCK) { + error = EINVAL; + break; + } + } + + if (cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) { + /* + * TBD OFD-style locking is currently limited to + * covering the entire file. + */ + if (bf.l_whence != 0 || bf.l_start != 0 || + bf.l_len != 0) { + error = EINVAL; + break; + } + } + + /* + * The *_frlock functions in the various file systems basically + * do some validation and then funnel everything through the + * fs_frlock function. For OFD-style locks fs_frlock will do + * nothing so that once control returns here we can call the + * ofdlock function with the correct fp. For OFD-style locks + * the unsupported remote file systems, such as NFS, detect and + * reject the OFD-style cmd argument. + */ if ((error = VOP_FRLOCK(vp, cmd, &bf, flag, offset, NULL, fp->f_cred, NULL)) != 0) break; - if ((cmd == F_GETLK) && bf.l_type == F_UNLCK) { + if (cmd == F_FLOCK || cmd == F_FLOCKW || cmd == F_OFD_GETLK || + cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) { + /* + * This is an OFD-style lock so we need to handle it + * here. Because OFD-style locks are associated with + * the file_t we didn't have enough info down the + * VOP_FRLOCK path immediately above. + */ + if ((error = ofdlock(fp, cmd, &bf, flag, offset)) != 0) + break; + } + + if ((cmd == F_GETLK || cmd == F_OFD_GETLK) && + bf.l_type == F_UNLCK) { if (copyout(&bf.l_type, &((struct flock *)arg)->l_type, sizeof (bf.l_type))) error = EFAULT; break; } - if (cmd == F_GETLK) { + if (cmd == F_GETLK || cmd == F_OFD_GETLK) { int i; /* diff --git a/usr/src/uts/common/syscall/memcntl.c b/usr/src/uts/common/syscall/memcntl.c index ae2a0cc45c..721f884a7e 100644 --- a/usr/src/uts/common/syscall/memcntl.c +++ b/usr/src/uts/common/syscall/memcntl.c @@ -115,13 +115,17 @@ memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg, int attr, int mask) * MS_SYNC used to be defined to be zero but is now non-zero. * For binary compatibility we still accept zero * (the absence of MS_ASYNC) to mean the same thing. + * Binary compatibility is not an issue for MS_INVALCURPROC. */ iarg = (uintptr_t)arg; if ((iarg & ~MS_INVALIDATE) == 0) iarg |= MS_SYNC; - if (((iarg & ~(MS_SYNC|MS_ASYNC|MS_INVALIDATE)) != 0) || - ((iarg & (MS_SYNC|MS_ASYNC)) == (MS_SYNC|MS_ASYNC))) { + if (((iarg & + ~(MS_SYNC|MS_ASYNC|MS_INVALIDATE|MS_INVALCURPROC)) != 0) || + ((iarg & (MS_SYNC|MS_ASYNC)) == (MS_SYNC|MS_ASYNC)) || + ((iarg & (MS_INVALIDATE|MS_INVALCURPROC)) == + (MS_INVALIDATE|MS_INVALCURPROC))) { error = set_errno(EINVAL); } else { error = as_ctl(as, addr, len, cmd, attr, iarg, NULL, 0); @@ -347,7 +351,8 @@ memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg, int attr, int mask) } return (error); case MC_ADVISE: - if ((uintptr_t)arg == MADV_FREE) { + if ((uintptr_t)arg == MADV_FREE || + (uintptr_t)arg == MADV_PURGE) { len &= PAGEMASK; } switch ((uintptr_t)arg) { diff --git a/usr/src/uts/common/syscall/open.c b/usr/src/uts/common/syscall/open.c index edb04c824b..874e31869c 100644 --- a/usr/src/uts/common/syscall/open.c +++ b/usr/src/uts/common/syscall/open.c @@ -74,12 +74,12 @@ copen(int startfd, char *fname, int filemode, int createmode) if (filemode & (FSEARCH|FEXEC)) { /* - * Must be one or the other and neither FREAD nor FWRITE + * Must be one or the other. * Must not be any of FAPPEND FCREAT FTRUNC FXATTR FXATTRDIROPEN - * XXX: Should these just be silently ignored? + * XXX: Should these just be silently ignored like we + * silently ignore FREAD|FWRITE? */ - if ((filemode & (FREAD|FWRITE)) || - (filemode & (FSEARCH|FEXEC)) == (FSEARCH|FEXEC) || + if ((filemode & (FSEARCH|FEXEC)) == (FSEARCH|FEXEC) || (filemode & (FAPPEND|FCREAT|FTRUNC|FXATTR|FXATTRDIROPEN))) return (set_errno(EINVAL)); } diff --git a/usr/src/uts/common/syscall/rusagesys.c b/usr/src/uts/common/syscall/rusagesys.c index 3e0e63f4c0..417c629168 100644 --- a/usr/src/uts/common/syscall/rusagesys.c +++ b/usr/src/uts/common/syscall/rusagesys.c @@ -21,6 +21,7 @@ /* * Copyright 2008 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. + * Copyright 2014 Joyent, Inc. All rights reserved. */ /* @@ -257,6 +258,19 @@ rusagesys(int code, void *arg1, void *arg2, void *arg3, void *arg4) case _RUSAGESYS_GETVMUSAGE: return (vm_getusage((uint_t)(uintptr_t)arg1, (time_t)arg2, (vmusage_t *)arg3, (size_t *)arg4, 0)); + case _RUSAGESYS_INVALMAP: + /* + * SPARC sfmmu hat does not support HAT_CURPROC_PGUNLOAD + * handling so callers on SPARC should get simple sync + * handling with invalidation to all processes. + */ +#if defined(__sparc) + return (memcntl((caddr_t)arg2, (size_t)arg3, MC_SYNC, + (caddr_t)(MS_ASYNC | MS_INVALIDATE), 0, 0)); +#else + return (vm_map_inval((pid_t)(uintptr_t)arg1, (caddr_t)arg2, + (size_t)arg3)); +#endif default: return (set_errno(EINVAL)); } diff --git a/usr/src/uts/common/syscall/rw.c b/usr/src/uts/common/syscall/rw.c index a28894b2c9..943b7d244e 100644 --- a/usr/src/uts/common/syscall/rw.c +++ b/usr/src/uts/common/syscall/rw.c @@ -22,7 +22,7 @@ /* * Copyright 2008 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. - * Copyright (c) 2015, Joyent, Inc. All rights reserved. + * Copyright 2015, Joyent, Inc. All rights reserved. */ /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */ @@ -50,6 +50,7 @@ #include <sys/debug.h> #include <sys/rctl.h> #include <sys/nbmlock.h> +#include <sys/limits.h> #define COPYOUT_MAX_CACHE (1<<17) /* 128K */ @@ -607,19 +608,12 @@ out: return (bcount); } -/* - * XXX -- The SVID refers to IOV_MAX, but doesn't define it. Grrrr.... - * XXX -- However, SVVS expects readv() and writev() to fail if - * XXX -- iovcnt > 16 (yes, it's hard-coded in the SVVS source), - * XXX -- so I guess that's the "interface". - */ -#define DEF_IOV_MAX 16 - ssize_t readv(int fdes, struct iovec *iovp, int iovcnt) { struct uio auio; - struct iovec aiov[DEF_IOV_MAX]; + struct iovec buf[IOV_MAX_STACK], *aiov = buf; + int aiovlen = 0; file_t *fp; register vnode_t *vp; struct cpu *cp; @@ -630,9 +624,14 @@ readv(int fdes, struct iovec *iovp, int iovcnt) u_offset_t fileoff; int in_crit = 0; - if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX) + if (iovcnt <= 0 || iovcnt > IOV_MAX) return (set_errno(EINVAL)); + if (iovcnt > IOV_MAX_STACK) { + aiovlen = iovcnt * sizeof (iovec_t); + aiov = kmem_alloc(aiovlen, KM_SLEEP); + } + #ifdef _SYSCALL32_IMPL /* * 32-bit callers need to have their iovec expanded, @@ -640,36 +639,63 @@ readv(int fdes, struct iovec *iovp, int iovcnt) * of data in a single call. */ if (get_udatamodel() == DATAMODEL_ILP32) { - struct iovec32 aiov32[DEF_IOV_MAX]; + struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32; + int aiov32len; ssize32_t count32; - if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32))) + aiov32len = iovcnt * sizeof (iovec32_t); + if (aiovlen != 0) + aiov32 = kmem_alloc(aiov32len, KM_SLEEP); + + if (copyin(iovp, aiov32, aiov32len)) { + if (aiovlen != 0) { + kmem_free(aiov32, aiov32len); + kmem_free(aiov, aiovlen); + } return (set_errno(EFAULT)); + } count32 = 0; for (i = 0; i < iovcnt; i++) { ssize32_t iovlen32 = aiov32[i].iov_len; count32 += iovlen32; - if (iovlen32 < 0 || count32 < 0) + if (iovlen32 < 0 || count32 < 0) { + if (aiovlen != 0) { + kmem_free(aiov32, aiov32len); + kmem_free(aiov, aiovlen); + } return (set_errno(EINVAL)); + } aiov[i].iov_len = iovlen32; aiov[i].iov_base = (caddr_t)(uintptr_t)aiov32[i].iov_base; } + + if (aiovlen != 0) + kmem_free(aiov32, aiov32len); } else #endif - if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec))) + if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EFAULT)); + } count = 0; for (i = 0; i < iovcnt; i++) { ssize_t iovlen = aiov[i].iov_len; count += iovlen; - if (iovlen < 0 || count < 0) + if (iovlen < 0 || count < 0) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EINVAL)); + } } - if ((fp = getf(fdes)) == NULL) + if ((fp = getf(fdes)) == NULL) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EBADF)); + } if (((fflag = fp->f_flag) & FREAD) == 0) { error = EBADF; goto out; @@ -768,6 +794,8 @@ out: if (in_crit) nbl_end_crit(vp); releasef(fdes); + if (aiovlen != 0) + kmem_free(aiov, aiovlen); if (error) return (set_errno(error)); return (count); @@ -777,7 +805,8 @@ ssize_t writev(int fdes, struct iovec *iovp, int iovcnt) { struct uio auio; - struct iovec aiov[DEF_IOV_MAX]; + struct iovec buf[IOV_MAX_STACK], *aiov = buf; + int aiovlen = 0; file_t *fp; register vnode_t *vp; struct cpu *cp; @@ -788,9 +817,14 @@ writev(int fdes, struct iovec *iovp, int iovcnt) u_offset_t fileoff; int in_crit = 0; - if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX) + if (iovcnt <= 0 || iovcnt > IOV_MAX) return (set_errno(EINVAL)); + if (iovcnt > IOV_MAX_STACK) { + aiovlen = iovcnt * sizeof (iovec_t); + aiov = kmem_alloc(aiovlen, KM_SLEEP); + } + #ifdef _SYSCALL32_IMPL /* * 32-bit callers need to have their iovec expanded, @@ -798,36 +832,62 @@ writev(int fdes, struct iovec *iovp, int iovcnt) * of data in a single call. */ if (get_udatamodel() == DATAMODEL_ILP32) { - struct iovec32 aiov32[DEF_IOV_MAX]; + struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32; + int aiov32len; ssize32_t count32; - if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32))) + aiov32len = iovcnt * sizeof (iovec32_t); + if (aiovlen != 0) + aiov32 = kmem_alloc(aiov32len, KM_SLEEP); + + if (copyin(iovp, aiov32, aiov32len)) { + if (aiovlen != 0) { + kmem_free(aiov32, aiov32len); + kmem_free(aiov, aiovlen); + } return (set_errno(EFAULT)); + } count32 = 0; for (i = 0; i < iovcnt; i++) { ssize32_t iovlen = aiov32[i].iov_len; count32 += iovlen; - if (iovlen < 0 || count32 < 0) + if (iovlen < 0 || count32 < 0) { + if (aiovlen != 0) { + kmem_free(aiov32, aiov32len); + kmem_free(aiov, aiovlen); + } return (set_errno(EINVAL)); + } aiov[i].iov_len = iovlen; aiov[i].iov_base = (caddr_t)(uintptr_t)aiov32[i].iov_base; } + if (aiovlen != 0) + kmem_free(aiov32, aiov32len); } else #endif - if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec))) + if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EFAULT)); + } count = 0; for (i = 0; i < iovcnt; i++) { ssize_t iovlen = aiov[i].iov_len; count += iovlen; - if (iovlen < 0 || count < 0) + if (iovlen < 0 || count < 0) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EINVAL)); + } } - if ((fp = getf(fdes)) == NULL) + if ((fp = getf(fdes)) == NULL) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EBADF)); + } if (((fflag = fp->f_flag) & FWRITE) == 0) { error = EBADF; goto out; @@ -917,6 +977,8 @@ out: if (in_crit) nbl_end_crit(vp); releasef(fdes); + if (aiovlen != 0) + kmem_free(aiov, aiovlen); if (error) return (set_errno(error)); return (count); @@ -927,7 +989,8 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset, off_t extended_offset) { struct uio auio; - struct iovec aiov[DEF_IOV_MAX]; + struct iovec buf[IOV_MAX_STACK], *aiov = buf; + int aiovlen = 0; file_t *fp; register vnode_t *vp; struct cpu *cp; @@ -952,9 +1015,14 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset, int in_crit = 0; - if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX) + if (iovcnt <= 0 || iovcnt > IOV_MAX) return (set_errno(EINVAL)); + if (iovcnt > IOV_MAX_STACK) { + aiovlen = iovcnt * sizeof (iovec_t); + aiov = kmem_alloc(aiovlen, KM_SLEEP); + } + #ifdef _SYSCALL32_IMPL /* * 32-bit callers need to have their iovec expanded, @@ -962,39 +1030,68 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset, * of data in a single call. */ if (get_udatamodel() == DATAMODEL_ILP32) { - struct iovec32 aiov32[DEF_IOV_MAX]; + struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32; + int aiov32len; ssize32_t count32; - if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32))) + aiov32len = iovcnt * sizeof (iovec32_t); + if (aiovlen != 0) + aiov32 = kmem_alloc(aiov32len, KM_SLEEP); + + if (copyin(iovp, aiov32, aiov32len)) { + if (aiovlen != 0) { + kmem_free(aiov32, aiov32len); + kmem_free(aiov, aiovlen); + } return (set_errno(EFAULT)); + } count32 = 0; for (i = 0; i < iovcnt; i++) { ssize32_t iovlen32 = aiov32[i].iov_len; count32 += iovlen32; - if (iovlen32 < 0 || count32 < 0) + if (iovlen32 < 0 || count32 < 0) { + if (aiovlen != 0) { + kmem_free(aiov32, aiov32len); + kmem_free(aiov, aiovlen); + } return (set_errno(EINVAL)); + } aiov[i].iov_len = iovlen32; aiov[i].iov_base = (caddr_t)(uintptr_t)aiov32[i].iov_base; } + if (aiovlen != 0) + kmem_free(aiov32, aiov32len); } else #endif /* _SYSCALL32_IMPL */ - if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec))) + if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EFAULT)); + } count = 0; for (i = 0; i < iovcnt; i++) { ssize_t iovlen = aiov[i].iov_len; count += iovlen; - if (iovlen < 0 || count < 0) + if (iovlen < 0 || count < 0) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EINVAL)); + } } - if ((bcount = (ssize_t)count) < 0) + if ((bcount = (ssize_t)count) < 0) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EINVAL)); - if ((fp = getf(fdes)) == NULL) + } + if ((fp = getf(fdes)) == NULL) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EBADF)); + } if (((fflag = fp->f_flag) & FREAD) == 0) { error = EBADF; goto out; @@ -1099,6 +1196,8 @@ out: if (in_crit) nbl_end_crit(vp); releasef(fdes); + if (aiovlen != 0) + kmem_free(aiov, aiovlen); if (error) return (set_errno(error)); return (count); @@ -1109,7 +1208,8 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset, off_t extended_offset) { struct uio auio; - struct iovec aiov[DEF_IOV_MAX]; + struct iovec buf[IOV_MAX_STACK], *aiov = buf; + int aiovlen = 0; file_t *fp; register vnode_t *vp; struct cpu *cp; @@ -1134,9 +1234,14 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset, int in_crit = 0; - if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX) + if (iovcnt <= 0 || iovcnt > IOV_MAX) return (set_errno(EINVAL)); + if (iovcnt > IOV_MAX_STACK) { + aiovlen = iovcnt * sizeof (iovec_t); + aiov = kmem_alloc(aiovlen, KM_SLEEP); + } + #ifdef _SYSCALL32_IMPL /* * 32-bit callers need to have their iovec expanded, @@ -1144,39 +1249,68 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset, * of data in a single call. */ if (get_udatamodel() == DATAMODEL_ILP32) { - struct iovec32 aiov32[DEF_IOV_MAX]; + struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32; + int aiov32len; ssize32_t count32; - if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32))) + aiov32len = iovcnt * sizeof (iovec32_t); + if (aiovlen != 0) + aiov32 = kmem_alloc(aiov32len, KM_SLEEP); + + if (copyin(iovp, aiov32, aiov32len)) { + if (aiovlen != 0) { + kmem_free(aiov32, aiov32len); + kmem_free(aiov, aiovlen); + } return (set_errno(EFAULT)); + } count32 = 0; for (i = 0; i < iovcnt; i++) { ssize32_t iovlen32 = aiov32[i].iov_len; count32 += iovlen32; - if (iovlen32 < 0 || count32 < 0) + if (iovlen32 < 0 || count32 < 0) { + if (aiovlen != 0) { + kmem_free(aiov32, aiov32len); + kmem_free(aiov, aiovlen); + } return (set_errno(EINVAL)); + } aiov[i].iov_len = iovlen32; aiov[i].iov_base = (caddr_t)(uintptr_t)aiov32[i].iov_base; } + if (aiovlen != 0) + kmem_free(aiov32, aiov32len); } else #endif /* _SYSCALL32_IMPL */ - if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec))) + if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EFAULT)); + } count = 0; for (i = 0; i < iovcnt; i++) { ssize_t iovlen = aiov[i].iov_len; count += iovlen; - if (iovlen < 0 || count < 0) + if (iovlen < 0 || count < 0) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EINVAL)); + } } - if ((bcount = (ssize_t)count) < 0) + if ((bcount = (ssize_t)count) < 0) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EINVAL)); - if ((fp = getf(fdes)) == NULL) + } + if ((fp = getf(fdes)) == NULL) { + if (aiovlen != 0) + kmem_free(aiov, aiovlen); return (set_errno(EBADF)); + } if (((fflag = fp->f_flag) & FWRITE) == 0) { error = EBADF; goto out; @@ -1308,6 +1442,8 @@ out: if (in_crit) nbl_end_crit(vp); releasef(fdes); + if (aiovlen != 0) + kmem_free(aiov, aiovlen); if (error) return (set_errno(error)); return (count); diff --git a/usr/src/uts/common/syscall/sendfile.c b/usr/src/uts/common/syscall/sendfile.c index cb8246f584..ccceca7c6d 100644 --- a/usr/src/uts/common/syscall/sendfile.c +++ b/usr/src/uts/common/syscall/sendfile.c @@ -82,7 +82,7 @@ extern sotpi_info_t *sotpi_sototpi(struct sonode *); * 64 bit kernel or 32 bit kernel. For 32 bit apps, we can't transfer * more than 2GB of data. */ -int +static int sendvec_chunk64(file_t *fp, u_offset_t *fileoff, struct ksendfilevec64 *sfv, int copy_cnt, ssize32_t *count) { @@ -343,7 +343,7 @@ sendvec_chunk64(file_t *fp, u_offset_t *fileoff, struct ksendfilevec64 *sfv, return (0); } -ssize32_t +static ssize32_t sendvec64(file_t *fp, const struct ksendfilevec64 *vec, int sfvcnt, size32_t *xferred, int fildes) { @@ -390,7 +390,7 @@ sendvec64(file_t *fp, const struct ksendfilevec64 *vec, int sfvcnt, } #endif -int +static int sendvec_small_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv, int copy_cnt, ssize_t total_size, int maxblk, ssize_t *count) { @@ -680,7 +680,7 @@ sendvec_small_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv, } -int +static int sendvec_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv, int copy_cnt, ssize_t *count) { @@ -1160,6 +1160,17 @@ sendfilev(int opcode, int fildes, const struct sendfilevec *vec, int sfvcnt, } else { maxblk = (int)vp->v_stream->sd_maxblk; } + + /* + * We need to make sure that the socket that we're sending on + * supports sendfile behavior. sockfs doesn't know that the APIs + * we want to use are coming from sendfile, so we can't rely on + * it to check for us. + */ + if ((so->so_mode & SM_SENDFILESUPP) == 0) { + error = EOPNOTSUPP; + goto err; + } break; case VREG: break; diff --git a/usr/src/uts/common/syscall/sysconfig.c b/usr/src/uts/common/syscall/sysconfig.c index 03f2fabe13..26ea859224 100644 --- a/usr/src/uts/common/syscall/sysconfig.c +++ b/usr/src/uts/common/syscall/sysconfig.c @@ -22,6 +22,7 @@ /* * Copyright 2008 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. + * Copyright 2012 Joyent, Inc. All rights reserved. */ /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */ @@ -170,8 +171,8 @@ sysconfig(int which) * even though rcapd can be used on the global zone too. */ if (!INGLOBALZONE(curproc) && - curproc->p_zone->zone_phys_mcap != 0) - return (MIN(btop(curproc->p_zone->zone_phys_mcap), + curproc->p_zone->zone_phys_mem_ctl != UINT64_MAX) + return (MIN(btop(curproc->p_zone->zone_phys_mem_ctl), physinstalled)); return (physinstalled); @@ -179,26 +180,23 @@ sysconfig(int which) case _CONFIG_AVPHYS_PAGES: /* * If the non-global zone has a phys. memory cap, use - * the phys. memory cap - zone's current rss. We always + * the phys. memory cap - zone's rss. We always * report the system-wide value for the global zone, even - * though rcapd can be used on the global zone too. + * though memory capping can be used on the global zone too. + * We use the cached value for the RSS since vm_getusage() + * is so expensive and we don't need this value to be exact. */ if (!INGLOBALZONE(curproc) && - curproc->p_zone->zone_phys_mcap != 0) { + curproc->p_zone->zone_phys_mem_ctl != UINT64_MAX) { pgcnt_t cap, rss, free; - vmusage_t in_use; - size_t cnt = 1; - cap = btop(curproc->p_zone->zone_phys_mcap); + cap = btop(curproc->p_zone->zone_phys_mem_ctl); if (cap > physinstalled) return (freemem); - if (vm_getusage(VMUSAGE_ZONE, 1, &in_use, &cnt, - FKIOCTL) != 0) - in_use.vmu_rss_all = 0; - rss = btop(in_use.vmu_rss_all); + rss = btop(curproc->p_zone->zone_phys_mem); /* - * Because rcapd implements a soft cap, it is possible + * Because this is a soft cap, it is possible * for rss to be temporarily over the cap. */ if (cap > rss) diff --git a/usr/src/uts/common/syscall/uadmin.c b/usr/src/uts/common/syscall/uadmin.c index 2dda4001bf..68aa1a95f5 100644 --- a/usr/src/uts/common/syscall/uadmin.c +++ b/usr/src/uts/common/syscall/uadmin.c @@ -78,7 +78,7 @@ volatile int fastreboot_dryrun = 0; * system with many zones. */ void -killall(zoneid_t zoneid) +killall(zoneid_t zoneid, boolean_t force) { proc_t *p; @@ -108,7 +108,7 @@ killall(zoneid_t zoneid) p->p_stat != SIDL && p->p_stat != SZOMB) { mutex_enter(&p->p_lock); - if (sigismember(&p->p_sig, SIGKILL)) { + if (!force && sigismember(&p->p_sig, SIGKILL)) { mutex_exit(&p->p_lock); p = p->p_next; } else { @@ -245,7 +245,7 @@ kadmin(int cmd, int fcn, void *mdep, cred_t *credp) */ zone_shutdown_global(); - killall(ALL_ZONES); + killall(ALL_ZONES, B_FALSE); /* * If we are calling kadmin() from a kernel context then we * do not release these resources. |