diff --git a/sys/compat/linuxkpi/common/src/linux_compat.c b/sys/compat/linuxkpi/common/src/linux_compat.c index dc8789741532..b73f6fed03af 100644 --- a/sys/compat/linuxkpi/common/src/linux_compat.c +++ b/sys/compat/linuxkpi/common/src/linux_compat.c @@ -1,2551 +1,2543 @@ /*- * Copyright (c) 2010 Isilon Systems, Inc. * Copyright (c) 2010 iX Systems, Inc. * Copyright (c) 2010 Panasas, Inc. * Copyright (c) 2013-2018 Mellanox Technologies, Ltd. * 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 unmodified, 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 ``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 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_stack.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 #if defined(__i386__) || defined(__amd64__) #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(__i386__) || defined(__amd64__) #include #endif SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "LinuxKPI parameters"); int linuxkpi_debug; SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN, &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable."); MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat"); #include /* Undo Linux compat changes. */ #undef RB_ROOT #undef file #undef cdev #define RB_ROOT(head) (head)->rbh_root static void linux_cdev_deref(struct linux_cdev *ldev); static struct vm_area_struct *linux_cdev_handle_find(void *handle); struct kobject linux_class_root; struct device linux_root_device; struct class linux_class_misc; struct list_head pci_drivers; struct list_head pci_devices; spinlock_t pci_lock; unsigned long linux_timer_hz_mask; wait_queue_head_t linux_bit_waitq; wait_queue_head_t linux_var_waitq; int panic_cmp(struct rb_node *one, struct rb_node *two) { panic("no cmp"); } RB_GENERATE(linux_root, rb_node, __entry, panic_cmp); int kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args) { va_list tmp_va; int len; char *old; char *name; char dummy; old = kobj->name; if (old && fmt == NULL) return (0); /* compute length of string */ va_copy(tmp_va, args); len = vsnprintf(&dummy, 0, fmt, tmp_va); va_end(tmp_va); /* account for zero termination */ len++; /* check for error */ if (len < 1) return (-EINVAL); /* allocate memory for string */ name = kzalloc(len, GFP_KERNEL); if (name == NULL) return (-ENOMEM); vsnprintf(name, len, fmt, args); kobj->name = name; /* free old string */ kfree(old); /* filter new string */ for (; *name != '\0'; name++) if (*name == '/') *name = '!'; return (0); } int kobject_set_name(struct kobject *kobj, const char *fmt, ...) { va_list args; int error; va_start(args, fmt); error = kobject_set_name_vargs(kobj, fmt, args); va_end(args); return (error); } static int kobject_add_complete(struct kobject *kobj, struct kobject *parent) { const struct kobj_type *t; int error; kobj->parent = parent; error = sysfs_create_dir(kobj); if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) { struct attribute **attr; t = kobj->ktype; for (attr = t->default_attrs; *attr != NULL; attr++) { error = sysfs_create_file(kobj, *attr); if (error) break; } if (error) sysfs_remove_dir(kobj); } return (error); } int kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...) { va_list args; int error; va_start(args, fmt); error = kobject_set_name_vargs(kobj, fmt, args); va_end(args); if (error) return (error); return kobject_add_complete(kobj, parent); } void linux_kobject_release(struct kref *kref) { struct kobject *kobj; char *name; kobj = container_of(kref, struct kobject, kref); sysfs_remove_dir(kobj); name = kobj->name; if (kobj->ktype && kobj->ktype->release) kobj->ktype->release(kobj); kfree(name); } static void linux_kobject_kfree(struct kobject *kobj) { kfree(kobj); } static void linux_kobject_kfree_name(struct kobject *kobj) { if (kobj) { kfree(kobj->name); } } const struct kobj_type linux_kfree_type = { .release = linux_kobject_kfree }; static void linux_device_release(struct device *dev) { pr_debug("linux_device_release: %s\n", dev_name(dev)); kfree(dev); } static ssize_t linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf) { struct class_attribute *dattr; ssize_t error; dattr = container_of(attr, struct class_attribute, attr); error = -EIO; if (dattr->show) error = dattr->show(container_of(kobj, struct class, kobj), dattr, buf); return (error); } static ssize_t linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf, size_t count) { struct class_attribute *dattr; ssize_t error; dattr = container_of(attr, struct class_attribute, attr); error = -EIO; if (dattr->store) error = dattr->store(container_of(kobj, struct class, kobj), dattr, buf, count); return (error); } static void linux_class_release(struct kobject *kobj) { struct class *class; class = container_of(kobj, struct class, kobj); if (class->class_release) class->class_release(class); } static const struct sysfs_ops linux_class_sysfs = { .show = linux_class_show, .store = linux_class_store, }; const struct kobj_type linux_class_ktype = { .release = linux_class_release, .sysfs_ops = &linux_class_sysfs }; static void linux_dev_release(struct kobject *kobj) { struct device *dev; dev = container_of(kobj, struct device, kobj); /* This is the precedence defined by linux. */ if (dev->release) dev->release(dev); else if (dev->class && dev->class->dev_release) dev->class->dev_release(dev); } static ssize_t linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf) { struct device_attribute *dattr; ssize_t error; dattr = container_of(attr, struct device_attribute, attr); error = -EIO; if (dattr->show) error = dattr->show(container_of(kobj, struct device, kobj), dattr, buf); return (error); } static ssize_t linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf, size_t count) { struct device_attribute *dattr; ssize_t error; dattr = container_of(attr, struct device_attribute, attr); error = -EIO; if (dattr->store) error = dattr->store(container_of(kobj, struct device, kobj), dattr, buf, count); return (error); } static const struct sysfs_ops linux_dev_sysfs = { .show = linux_dev_show, .store = linux_dev_store, }; const struct kobj_type linux_dev_ktype = { .release = linux_dev_release, .sysfs_ops = &linux_dev_sysfs }; struct device * device_create(struct class *class, struct device *parent, dev_t devt, void *drvdata, const char *fmt, ...) { struct device *dev; va_list args; dev = kzalloc(sizeof(*dev), M_WAITOK); dev->parent = parent; dev->class = class; dev->devt = devt; dev->driver_data = drvdata; dev->release = linux_device_release; va_start(args, fmt); kobject_set_name_vargs(&dev->kobj, fmt, args); va_end(args); device_register(dev); return (dev); } int kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype, struct kobject *parent, const char *fmt, ...) { va_list args; int error; kobject_init(kobj, ktype); kobj->ktype = ktype; kobj->parent = parent; kobj->name = NULL; va_start(args, fmt); error = kobject_set_name_vargs(kobj, fmt, args); va_end(args); if (error) return (error); return kobject_add_complete(kobj, parent); } static void linux_kq_lock(void *arg) { spinlock_t *s = arg; spin_lock(s); } static void linux_kq_unlock(void *arg) { spinlock_t *s = arg; spin_unlock(s); } static void -linux_kq_lock_owned(void *arg) +linux_kq_assert_lock(void *arg, int what) { #ifdef INVARIANTS spinlock_t *s = arg; - mtx_assert(&s->m, MA_OWNED); -#endif -} - -static void -linux_kq_lock_unowned(void *arg) -{ -#ifdef INVARIANTS - spinlock_t *s = arg; - - mtx_assert(&s->m, MA_NOTOWNED); + if (what == LA_LOCKED) + mtx_assert(&s->m, MA_OWNED); + else + mtx_assert(&s->m, MA_NOTOWNED); #endif } static void linux_file_kqfilter_poll(struct linux_file *, int); struct linux_file * linux_file_alloc(void) { struct linux_file *filp; filp = kzalloc(sizeof(*filp), GFP_KERNEL); /* set initial refcount */ filp->f_count = 1; /* setup fields needed by kqueue support */ spin_lock_init(&filp->f_kqlock); knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock, - linux_kq_lock, linux_kq_unlock, - linux_kq_lock_owned, linux_kq_lock_unowned); + linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock); return (filp); } void linux_file_free(struct linux_file *filp) { if (filp->_file == NULL) { if (filp->f_shmem != NULL) vm_object_deallocate(filp->f_shmem); kfree(filp); } else { /* * The close method of the character device or file * will free the linux_file structure: */ _fdrop(filp->_file, curthread); } } static int linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot, vm_page_t *mres) { struct vm_area_struct *vmap; vmap = linux_cdev_handle_find(vm_obj->handle); MPASS(vmap != NULL); MPASS(vmap->vm_private_data == vm_obj->handle); if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) { vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset; vm_page_t page; if (((*mres)->flags & PG_FICTITIOUS) != 0) { /* * If the passed in result page is a fake * page, update it with the new physical * address. */ page = *mres; vm_page_updatefake(page, paddr, vm_obj->memattr); } else { /* * Replace the passed in "mres" page with our * own fake page and free up the all of the * original pages. */ VM_OBJECT_WUNLOCK(vm_obj); page = vm_page_getfake(paddr, vm_obj->memattr); VM_OBJECT_WLOCK(vm_obj); vm_page_replace(page, vm_obj, (*mres)->pindex, *mres); *mres = page; } vm_page_valid(page); return (VM_PAGER_OK); } return (VM_PAGER_FAIL); } static int linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type, vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last) { struct vm_area_struct *vmap; int err; /* get VM area structure */ vmap = linux_cdev_handle_find(vm_obj->handle); MPASS(vmap != NULL); MPASS(vmap->vm_private_data == vm_obj->handle); VM_OBJECT_WUNLOCK(vm_obj); linux_set_current(curthread); down_write(&vmap->vm_mm->mmap_sem); if (unlikely(vmap->vm_ops == NULL)) { err = VM_FAULT_SIGBUS; } else { struct vm_fault vmf; /* fill out VM fault structure */ vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx); vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0; vmf.pgoff = 0; vmf.page = NULL; vmf.vma = vmap; vmap->vm_pfn_count = 0; vmap->vm_pfn_pcount = &vmap->vm_pfn_count; vmap->vm_obj = vm_obj; err = vmap->vm_ops->fault(vmap, &vmf); while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) { kern_yield(PRI_USER); err = vmap->vm_ops->fault(vmap, &vmf); } } /* translate return code */ switch (err) { case VM_FAULT_OOM: err = VM_PAGER_AGAIN; break; case VM_FAULT_SIGBUS: err = VM_PAGER_BAD; break; case VM_FAULT_NOPAGE: /* * By contract the fault handler will return having * busied all the pages itself. If pidx is already * found in the object, it will simply xbusy the first * page and return with vm_pfn_count set to 1. */ *first = vmap->vm_pfn_first; *last = *first + vmap->vm_pfn_count - 1; err = VM_PAGER_OK; break; default: err = VM_PAGER_ERROR; break; } up_write(&vmap->vm_mm->mmap_sem); VM_OBJECT_WLOCK(vm_obj); return (err); } static struct rwlock linux_vma_lock; static TAILQ_HEAD(, vm_area_struct) linux_vma_head = TAILQ_HEAD_INITIALIZER(linux_vma_head); static void linux_cdev_handle_free(struct vm_area_struct *vmap) { /* Drop reference on vm_file */ if (vmap->vm_file != NULL) fput(vmap->vm_file); /* Drop reference on mm_struct */ mmput(vmap->vm_mm); kfree(vmap); } static void linux_cdev_handle_remove(struct vm_area_struct *vmap) { rw_wlock(&linux_vma_lock); TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry); rw_wunlock(&linux_vma_lock); } static struct vm_area_struct * linux_cdev_handle_find(void *handle) { struct vm_area_struct *vmap; rw_rlock(&linux_vma_lock); TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) { if (vmap->vm_private_data == handle) break; } rw_runlock(&linux_vma_lock); return (vmap); } static int linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred, u_short *color) { MPASS(linux_cdev_handle_find(handle) != NULL); *color = 0; return (0); } static void linux_cdev_pager_dtor(void *handle) { const struct vm_operations_struct *vm_ops; struct vm_area_struct *vmap; vmap = linux_cdev_handle_find(handle); MPASS(vmap != NULL); /* * Remove handle before calling close operation to prevent * other threads from reusing the handle pointer. */ linux_cdev_handle_remove(vmap); down_write(&vmap->vm_mm->mmap_sem); vm_ops = vmap->vm_ops; if (likely(vm_ops != NULL)) vm_ops->close(vmap); up_write(&vmap->vm_mm->mmap_sem); linux_cdev_handle_free(vmap); } static struct cdev_pager_ops linux_cdev_pager_ops[2] = { { /* OBJT_MGTDEVICE */ .cdev_pg_populate = linux_cdev_pager_populate, .cdev_pg_ctor = linux_cdev_pager_ctor, .cdev_pg_dtor = linux_cdev_pager_dtor }, { /* OBJT_DEVICE */ .cdev_pg_fault = linux_cdev_pager_fault, .cdev_pg_ctor = linux_cdev_pager_ctor, .cdev_pg_dtor = linux_cdev_pager_dtor }, }; int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, unsigned long size) { vm_object_t obj; vm_page_t m; obj = vma->vm_obj; if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0) return (-ENOTSUP); VM_OBJECT_RLOCK(obj); for (m = vm_page_find_least(obj, OFF_TO_IDX(address)); m != NULL && m->pindex < OFF_TO_IDX(address + size); m = TAILQ_NEXT(m, listq)) pmap_remove_all(m); VM_OBJECT_RUNLOCK(obj); return (0); } static struct file_operations dummy_ldev_ops = { /* XXXKIB */ }; static struct linux_cdev dummy_ldev = { .ops = &dummy_ldev_ops, }; #define LDEV_SI_DTR 0x0001 #define LDEV_SI_REF 0x0002 static void linux_get_fop(struct linux_file *filp, const struct file_operations **fop, struct linux_cdev **dev) { struct linux_cdev *ldev; u_int siref; ldev = filp->f_cdev; *fop = filp->f_op; if (ldev != NULL) { for (siref = ldev->siref;;) { if ((siref & LDEV_SI_DTR) != 0) { ldev = &dummy_ldev; siref = ldev->siref; *fop = ldev->ops; MPASS((ldev->siref & LDEV_SI_DTR) == 0); } else if (atomic_fcmpset_int(&ldev->siref, &siref, siref + LDEV_SI_REF)) { break; } } } *dev = ldev; } static void linux_drop_fop(struct linux_cdev *ldev) { if (ldev == NULL) return; MPASS((ldev->siref & ~LDEV_SI_DTR) != 0); atomic_subtract_int(&ldev->siref, LDEV_SI_REF); } #define OPW(fp,td,code) ({ \ struct file *__fpop; \ __typeof(code) __retval; \ \ __fpop = (td)->td_fpop; \ (td)->td_fpop = (fp); \ __retval = (code); \ (td)->td_fpop = __fpop; \ __retval; \ }) static int linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td, struct file *file) { struct linux_cdev *ldev; struct linux_file *filp; const struct file_operations *fop; int error; ldev = dev->si_drv1; filp = linux_file_alloc(); filp->f_dentry = &filp->f_dentry_store; filp->f_op = ldev->ops; filp->f_mode = file->f_flag; filp->f_flags = file->f_flag; filp->f_vnode = file->f_vnode; filp->_file = file; refcount_acquire(&ldev->refs); filp->f_cdev = ldev; linux_set_current(td); linux_get_fop(filp, &fop, &ldev); if (fop->open != NULL) { error = -fop->open(file->f_vnode, filp); if (error != 0) { linux_drop_fop(ldev); linux_cdev_deref(filp->f_cdev); kfree(filp); return (error); } } /* hold on to the vnode - used for fstat() */ vhold(filp->f_vnode); /* release the file from devfs */ finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops); linux_drop_fop(ldev); return (ENXIO); } #define LINUX_IOCTL_MIN_PTR 0x10000UL #define LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX) static inline int linux_remap_address(void **uaddr, size_t len) { uintptr_t uaddr_val = (uintptr_t)(*uaddr); if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR && uaddr_val < LINUX_IOCTL_MAX_PTR)) { struct task_struct *pts = current; if (pts == NULL) { *uaddr = NULL; return (1); } /* compute data offset */ uaddr_val -= LINUX_IOCTL_MIN_PTR; /* check that length is within bounds */ if ((len > IOCPARM_MAX) || (uaddr_val + len) > pts->bsd_ioctl_len) { *uaddr = NULL; return (1); } /* re-add kernel buffer address */ uaddr_val += (uintptr_t)pts->bsd_ioctl_data; /* update address location */ *uaddr = (void *)uaddr_val; return (1); } return (0); } int linux_copyin(const void *uaddr, void *kaddr, size_t len) { if (linux_remap_address(__DECONST(void **, &uaddr), len)) { if (uaddr == NULL) return (-EFAULT); memcpy(kaddr, uaddr, len); return (0); } return (-copyin(uaddr, kaddr, len)); } int linux_copyout(const void *kaddr, void *uaddr, size_t len) { if (linux_remap_address(&uaddr, len)) { if (uaddr == NULL) return (-EFAULT); memcpy(uaddr, kaddr, len); return (0); } return (-copyout(kaddr, uaddr, len)); } size_t linux_clear_user(void *_uaddr, size_t _len) { uint8_t *uaddr = _uaddr; size_t len = _len; /* make sure uaddr is aligned before going into the fast loop */ while (((uintptr_t)uaddr & 7) != 0 && len > 7) { if (subyte(uaddr, 0)) return (_len); uaddr++; len--; } /* zero 8 bytes at a time */ while (len > 7) { #ifdef __LP64__ if (suword64(uaddr, 0)) return (_len); #else if (suword32(uaddr, 0)) return (_len); if (suword32(uaddr + 4, 0)) return (_len); #endif uaddr += 8; len -= 8; } /* zero fill end, if any */ while (len > 0) { if (subyte(uaddr, 0)) return (_len); uaddr++; len--; } return (0); } int linux_access_ok(const void *uaddr, size_t len) { uintptr_t saddr; uintptr_t eaddr; /* get start and end address */ saddr = (uintptr_t)uaddr; eaddr = (uintptr_t)uaddr + len; /* verify addresses are valid for userspace */ return ((saddr == eaddr) || (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS)); } /* * This function should return either EINTR or ERESTART depending on * the signal type sent to this thread: */ static int linux_get_error(struct task_struct *task, int error) { /* check for signal type interrupt code */ if (error == EINTR || error == ERESTARTSYS || error == ERESTART) { error = -linux_schedule_get_interrupt_value(task); if (error == 0) error = EINTR; } return (error); } static int linux_file_ioctl_sub(struct file *fp, struct linux_file *filp, const struct file_operations *fop, u_long cmd, caddr_t data, struct thread *td) { struct task_struct *task = current; unsigned size; int error; size = IOCPARM_LEN(cmd); /* refer to logic in sys_ioctl() */ if (size > 0) { /* * Setup hint for linux_copyin() and linux_copyout(). * * Background: Linux code expects a user-space address * while FreeBSD supplies a kernel-space address. */ task->bsd_ioctl_data = data; task->bsd_ioctl_len = size; data = (void *)LINUX_IOCTL_MIN_PTR; } else { /* fetch user-space pointer */ data = *(void **)data; } #if defined(__amd64__) if (td->td_proc->p_elf_machine == EM_386) { /* try the compat IOCTL handler first */ if (fop->compat_ioctl != NULL) { error = -OPW(fp, td, fop->compat_ioctl(filp, cmd, (u_long)data)); } else { error = ENOTTY; } /* fallback to the regular IOCTL handler, if any */ if (error == ENOTTY && fop->unlocked_ioctl != NULL) { error = -OPW(fp, td, fop->unlocked_ioctl(filp, cmd, (u_long)data)); } } else #endif { if (fop->unlocked_ioctl != NULL) { error = -OPW(fp, td, fop->unlocked_ioctl(filp, cmd, (u_long)data)); } else { error = ENOTTY; } } if (size > 0) { task->bsd_ioctl_data = NULL; task->bsd_ioctl_len = 0; } if (error == EWOULDBLOCK) { /* update kqfilter status, if any */ linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); } else { error = linux_get_error(task, error); } return (error); } #define LINUX_POLL_TABLE_NORMAL ((poll_table *)1) /* * This function atomically updates the poll wakeup state and returns * the previous state at the time of update. */ static uint8_t linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate) { int c, old; c = v->counter; while ((old = atomic_cmpxchg(v, c, pstate[c])) != c) c = old; return (c); } static int linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key) { static const uint8_t state[LINUX_FWQ_STATE_MAX] = { [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY, [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */ }; struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq); switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { case LINUX_FWQ_STATE_QUEUED: linux_poll_wakeup(filp); return (1); default: return (0); } } void linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p) { static const uint8_t state[LINUX_FWQ_STATE_MAX] = { [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY, [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */ [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED, }; /* check if we are called inside the select system call */ if (p == LINUX_POLL_TABLE_NORMAL) selrecord(curthread, &filp->f_selinfo); switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { case LINUX_FWQ_STATE_INIT: /* NOTE: file handles can only belong to one wait-queue */ filp->f_wait_queue.wqh = wqh; filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback; add_wait_queue(wqh, &filp->f_wait_queue.wq); atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED); break; default: break; } } static void linux_poll_wait_dequeue(struct linux_file *filp) { static const uint8_t state[LINUX_FWQ_STATE_MAX] = { [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT, [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT, [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT, }; seldrain(&filp->f_selinfo); switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { case LINUX_FWQ_STATE_NOT_READY: case LINUX_FWQ_STATE_QUEUED: case LINUX_FWQ_STATE_READY: remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq); break; default: break; } } void linux_poll_wakeup(struct linux_file *filp) { /* this function should be NULL-safe */ if (filp == NULL) return; selwakeup(&filp->f_selinfo); spin_lock(&filp->f_kqlock); filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ | LINUX_KQ_FLAG_NEED_WRITE; /* make sure the "knote" gets woken up */ KNOTE_LOCKED(&filp->f_selinfo.si_note, 1); spin_unlock(&filp->f_kqlock); } static void linux_file_kqfilter_detach(struct knote *kn) { struct linux_file *filp = kn->kn_hook; spin_lock(&filp->f_kqlock); knlist_remove(&filp->f_selinfo.si_note, kn, 1); spin_unlock(&filp->f_kqlock); } static int linux_file_kqfilter_read_event(struct knote *kn, long hint) { struct linux_file *filp = kn->kn_hook; mtx_assert(&filp->f_kqlock.m, MA_OWNED); return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0); } static int linux_file_kqfilter_write_event(struct knote *kn, long hint) { struct linux_file *filp = kn->kn_hook; mtx_assert(&filp->f_kqlock.m, MA_OWNED); return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0); } static struct filterops linux_dev_kqfiltops_read = { .f_isfd = 1, .f_detach = linux_file_kqfilter_detach, .f_event = linux_file_kqfilter_read_event, }; static struct filterops linux_dev_kqfiltops_write = { .f_isfd = 1, .f_detach = linux_file_kqfilter_detach, .f_event = linux_file_kqfilter_write_event, }; static void linux_file_kqfilter_poll(struct linux_file *filp, int kqflags) { struct thread *td; const struct file_operations *fop; struct linux_cdev *ldev; int temp; if ((filp->f_kqflags & kqflags) == 0) return; td = curthread; linux_get_fop(filp, &fop, &ldev); /* get the latest polling state */ temp = OPW(filp->_file, td, fop->poll(filp, NULL)); linux_drop_fop(ldev); spin_lock(&filp->f_kqlock); /* clear kqflags */ filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ | LINUX_KQ_FLAG_NEED_WRITE); /* update kqflags */ if ((temp & (POLLIN | POLLOUT)) != 0) { if ((temp & POLLIN) != 0) filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ; if ((temp & POLLOUT) != 0) filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE; /* make sure the "knote" gets woken up */ KNOTE_LOCKED(&filp->f_selinfo.si_note, 0); } spin_unlock(&filp->f_kqlock); } static int linux_file_kqfilter(struct file *file, struct knote *kn) { struct linux_file *filp; struct thread *td; int error; td = curthread; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; if (filp->f_op->poll == NULL) return (EINVAL); spin_lock(&filp->f_kqlock); switch (kn->kn_filter) { case EVFILT_READ: filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ; kn->kn_fop = &linux_dev_kqfiltops_read; kn->kn_hook = filp; knlist_add(&filp->f_selinfo.si_note, kn, 1); error = 0; break; case EVFILT_WRITE: filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE; kn->kn_fop = &linux_dev_kqfiltops_write; kn->kn_hook = filp; knlist_add(&filp->f_selinfo.si_note, kn, 1); error = 0; break; default: error = EINVAL; break; } spin_unlock(&filp->f_kqlock); if (error == 0) { linux_set_current(td); /* update kqfilter status, if any */ linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); } return (error); } static int linux_file_mmap_single(struct file *fp, const struct file_operations *fop, vm_ooffset_t *offset, vm_size_t size, struct vm_object **object, int nprot, struct thread *td) { struct task_struct *task; struct vm_area_struct *vmap; struct mm_struct *mm; struct linux_file *filp; vm_memattr_t attr; int error; filp = (struct linux_file *)fp->f_data; filp->f_flags = fp->f_flag; if (fop->mmap == NULL) return (EOPNOTSUPP); linux_set_current(td); /* * The same VM object might be shared by multiple processes * and the mm_struct is usually freed when a process exits. * * The atomic reference below makes sure the mm_struct is * available as long as the vmap is in the linux_vma_head. */ task = current; mm = task->mm; if (atomic_inc_not_zero(&mm->mm_users) == 0) return (EINVAL); vmap = kzalloc(sizeof(*vmap), GFP_KERNEL); vmap->vm_start = 0; vmap->vm_end = size; vmap->vm_pgoff = *offset / PAGE_SIZE; vmap->vm_pfn = 0; vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL); vmap->vm_ops = NULL; vmap->vm_file = get_file(filp); vmap->vm_mm = mm; if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) { error = linux_get_error(task, EINTR); } else { error = -OPW(fp, td, fop->mmap(filp, vmap)); error = linux_get_error(task, error); up_write(&vmap->vm_mm->mmap_sem); } if (error != 0) { linux_cdev_handle_free(vmap); return (error); } attr = pgprot2cachemode(vmap->vm_page_prot); if (vmap->vm_ops != NULL) { struct vm_area_struct *ptr; void *vm_private_data; bool vm_no_fault; if (vmap->vm_ops->open == NULL || vmap->vm_ops->close == NULL || vmap->vm_private_data == NULL) { /* free allocated VM area struct */ linux_cdev_handle_free(vmap); return (EINVAL); } vm_private_data = vmap->vm_private_data; rw_wlock(&linux_vma_lock); TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) { if (ptr->vm_private_data == vm_private_data) break; } /* check if there is an existing VM area struct */ if (ptr != NULL) { /* check if the VM area structure is invalid */ if (ptr->vm_ops == NULL || ptr->vm_ops->open == NULL || ptr->vm_ops->close == NULL) { error = ESTALE; vm_no_fault = 1; } else { error = EEXIST; vm_no_fault = (ptr->vm_ops->fault == NULL); } } else { /* insert VM area structure into list */ TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry); error = 0; vm_no_fault = (vmap->vm_ops->fault == NULL); } rw_wunlock(&linux_vma_lock); if (error != 0) { /* free allocated VM area struct */ linux_cdev_handle_free(vmap); /* check for stale VM area struct */ if (error != EEXIST) return (error); } /* check if there is no fault handler */ if (vm_no_fault) { *object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE, &linux_cdev_pager_ops[1], size, nprot, *offset, td->td_ucred); } else { *object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE, &linux_cdev_pager_ops[0], size, nprot, *offset, td->td_ucred); } /* check if allocating the VM object failed */ if (*object == NULL) { if (error == 0) { /* remove VM area struct from list */ linux_cdev_handle_remove(vmap); /* free allocated VM area struct */ linux_cdev_handle_free(vmap); } return (EINVAL); } } else { struct sglist *sg; sg = sglist_alloc(1, M_WAITOK); sglist_append_phys(sg, (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len); *object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len, nprot, 0, td->td_ucred); linux_cdev_handle_free(vmap); if (*object == NULL) { sglist_free(sg); return (EINVAL); } } if (attr != VM_MEMATTR_DEFAULT) { VM_OBJECT_WLOCK(*object); vm_object_set_memattr(*object, attr); VM_OBJECT_WUNLOCK(*object); } *offset = 0; return (0); } struct cdevsw linuxcdevsw = { .d_version = D_VERSION, .d_fdopen = linux_dev_fdopen, .d_name = "lkpidev", }; static int linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; ssize_t bytes; int error; error = 0; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; /* XXX no support for I/O vectors currently */ if (uio->uio_iovcnt != 1) return (EOPNOTSUPP); if (uio->uio_resid > DEVFS_IOSIZE_MAX) return (EINVAL); linux_set_current(td); linux_get_fop(filp, &fop, &ldev); if (fop->read != NULL) { bytes = OPW(file, td, fop->read(filp, uio->uio_iov->iov_base, uio->uio_iov->iov_len, &uio->uio_offset)); if (bytes >= 0) { uio->uio_iov->iov_base = ((uint8_t *)uio->uio_iov->iov_base) + bytes; uio->uio_iov->iov_len -= bytes; uio->uio_resid -= bytes; } else { error = linux_get_error(current, -bytes); } } else error = ENXIO; /* update kqfilter status, if any */ linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ); linux_drop_fop(ldev); return (error); } static int linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; ssize_t bytes; int error; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; /* XXX no support for I/O vectors currently */ if (uio->uio_iovcnt != 1) return (EOPNOTSUPP); if (uio->uio_resid > DEVFS_IOSIZE_MAX) return (EINVAL); linux_set_current(td); linux_get_fop(filp, &fop, &ldev); if (fop->write != NULL) { bytes = OPW(file, td, fop->write(filp, uio->uio_iov->iov_base, uio->uio_iov->iov_len, &uio->uio_offset)); if (bytes >= 0) { uio->uio_iov->iov_base = ((uint8_t *)uio->uio_iov->iov_base) + bytes; uio->uio_iov->iov_len -= bytes; uio->uio_resid -= bytes; error = 0; } else { error = linux_get_error(current, -bytes); } } else error = ENXIO; /* update kqfilter status, if any */ linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE); linux_drop_fop(ldev); return (error); } static int linux_file_poll(struct file *file, int events, struct ucred *active_cred, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; int revents; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; linux_set_current(td); linux_get_fop(filp, &fop, &ldev); if (fop->poll != NULL) { revents = OPW(file, td, fop->poll(filp, LINUX_POLL_TABLE_NORMAL)) & events; } else { revents = 0; } linux_drop_fop(ldev); return (revents); } static int linux_file_close(struct file *file, struct thread *td) { struct linux_file *filp; int (*release)(struct inode *, struct linux_file *); const struct file_operations *fop; struct linux_cdev *ldev; int error; filp = (struct linux_file *)file->f_data; KASSERT(file_count(filp) == 0, ("File refcount(%d) is not zero", file_count(filp))); if (td == NULL) td = curthread; error = 0; filp->f_flags = file->f_flag; linux_set_current(td); linux_poll_wait_dequeue(filp); linux_get_fop(filp, &fop, &ldev); /* * Always use the real release function, if any, to avoid * leaking device resources: */ release = filp->f_op->release; if (release != NULL) error = -OPW(file, td, release(filp->f_vnode, filp)); funsetown(&filp->f_sigio); if (filp->f_vnode != NULL) vdrop(filp->f_vnode); linux_drop_fop(ldev); if (filp->f_cdev != NULL) linux_cdev_deref(filp->f_cdev); kfree(filp); return (error); } static int linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; struct fiodgname_arg *fgn; const char *p; int error, i; error = 0; filp = (struct linux_file *)fp->f_data; filp->f_flags = fp->f_flag; linux_get_fop(filp, &fop, &ldev); linux_set_current(td); switch (cmd) { case FIONBIO: break; case FIOASYNC: if (fop->fasync == NULL) break; error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC)); break; case FIOSETOWN: error = fsetown(*(int *)data, &filp->f_sigio); if (error == 0) { if (fop->fasync == NULL) break; error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC)); } break; case FIOGETOWN: *(int *)data = fgetown(&filp->f_sigio); break; case FIODGNAME: #ifdef COMPAT_FREEBSD32 case FIODGNAME_32: #endif if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) { error = ENXIO; break; } fgn = data; p = devtoname(filp->f_cdev->cdev); i = strlen(p) + 1; if (i > fgn->len) { error = EINVAL; break; } error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i); break; default: error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td); break; } linux_drop_fop(ldev); return (error); } static int linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, struct file *fp, vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp) { /* * Character devices do not provide private mappings * of any kind: */ if ((*maxprotp & VM_PROT_WRITE) == 0 && (prot & VM_PROT_WRITE) != 0) return (EACCES); if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0) return (EINVAL); return (linux_file_mmap_single(fp, fop, foff, objsize, objp, (int)prot, td)); } static int linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; struct mount *mp; struct vnode *vp; vm_object_t object; vm_prot_t maxprot; int error; filp = (struct linux_file *)fp->f_data; vp = filp->f_vnode; if (vp == NULL) return (EOPNOTSUPP); /* * Ensure that file and memory protections are * compatible. */ mp = vp->v_mount; if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { maxprot = VM_PROT_NONE; if ((prot & VM_PROT_EXECUTE) != 0) return (EACCES); } else maxprot = VM_PROT_EXECUTE; if ((fp->f_flag & FREAD) != 0) maxprot |= VM_PROT_READ; else if ((prot & VM_PROT_READ) != 0) return (EACCES); /* * If we are sharing potential changes via MAP_SHARED and we * are trying to get write permission although we opened it * without asking for it, bail out. * * Note that most character devices always share mappings. * * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE * requests rather than doing it here. */ if ((flags & MAP_SHARED) != 0) { if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; else if ((prot & VM_PROT_WRITE) != 0) return (EACCES); } maxprot &= cap_maxprot; linux_get_fop(filp, &fop, &ldev); error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp, &foff, fop, &object); if (error != 0) goto out; error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, FALSE, td); if (error != 0) vm_object_deallocate(object); out: linux_drop_fop(ldev); return (error); } static int linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { struct linux_file *filp; struct vnode *vp; int error; filp = (struct linux_file *)fp->f_data; if (filp->f_vnode == NULL) return (EOPNOTSUPP); vp = filp->f_vnode; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_STAT(vp, sb, td->td_ucred, NOCRED, td); VOP_UNLOCK(vp); return (error); } static int linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct linux_file *filp; struct vnode *vp; int error; filp = fp->f_data; vp = filp->f_vnode; if (vp == NULL) { error = 0; kif->kf_type = KF_TYPE_DEV; } else { vref(vp); FILEDESC_SUNLOCK(fdp); error = vn_fill_kinfo_vnode(vp, kif); vrele(vp); kif->kf_type = KF_TYPE_VNODE; FILEDESC_SLOCK(fdp); } return (error); } unsigned int linux_iminor(struct inode *inode) { struct linux_cdev *ldev; if (inode == NULL || inode->v_rdev == NULL || inode->v_rdev->si_devsw != &linuxcdevsw) return (-1U); ldev = inode->v_rdev->si_drv1; if (ldev == NULL) return (-1U); return (minor(ldev->dev)); } struct fileops linuxfileops = { .fo_read = linux_file_read, .fo_write = linux_file_write, .fo_truncate = invfo_truncate, .fo_kqfilter = linux_file_kqfilter, .fo_stat = linux_file_stat, .fo_fill_kinfo = linux_file_fill_kinfo, .fo_poll = linux_file_poll, .fo_close = linux_file_close, .fo_ioctl = linux_file_ioctl, .fo_mmap = linux_file_mmap, .fo_chmod = invfo_chmod, .fo_chown = invfo_chown, .fo_sendfile = invfo_sendfile, .fo_flags = DFLAG_PASSABLE, }; /* * Hash of vmmap addresses. This is infrequently accessed and does not * need to be particularly large. This is done because we must store the * caller's idea of the map size to properly unmap. */ struct vmmap { LIST_ENTRY(vmmap) vm_next; void *vm_addr; unsigned long vm_size; }; struct vmmaphd { struct vmmap *lh_first; }; #define VMMAP_HASH_SIZE 64 #define VMMAP_HASH_MASK (VMMAP_HASH_SIZE - 1) #define VM_HASH(addr) ((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE]; static struct mtx vmmaplock; static void vmmap_add(void *addr, unsigned long size) { struct vmmap *vmmap; vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL); mtx_lock(&vmmaplock); vmmap->vm_size = size; vmmap->vm_addr = addr; LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next); mtx_unlock(&vmmaplock); } static struct vmmap * vmmap_remove(void *addr) { struct vmmap *vmmap; mtx_lock(&vmmaplock); LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next) if (vmmap->vm_addr == addr) break; if (vmmap) LIST_REMOVE(vmmap, vm_next); mtx_unlock(&vmmaplock); return (vmmap); } #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) void * _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr) { void *addr; addr = pmap_mapdev_attr(phys_addr, size, attr); if (addr == NULL) return (NULL); vmmap_add(addr, size); return (addr); } #endif void iounmap(void *addr) { struct vmmap *vmmap; vmmap = vmmap_remove(addr); if (vmmap == NULL) return; #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size); #endif kfree(vmmap); } void * vmap(struct page **pages, unsigned int count, unsigned long flags, int prot) { vm_offset_t off; size_t size; size = count * PAGE_SIZE; off = kva_alloc(size); if (off == 0) return (NULL); vmmap_add((void *)off, size); pmap_qenter(off, pages, count); return ((void *)off); } void vunmap(void *addr) { struct vmmap *vmmap; vmmap = vmmap_remove(addr); if (vmmap == NULL) return; pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE); kva_free((vm_offset_t)addr, vmmap->vm_size); kfree(vmmap); } char * kvasprintf(gfp_t gfp, const char *fmt, va_list ap) { unsigned int len; char *p; va_list aq; va_copy(aq, ap); len = vsnprintf(NULL, 0, fmt, aq); va_end(aq); p = kmalloc(len + 1, gfp); if (p != NULL) vsnprintf(p, len + 1, fmt, ap); return (p); } char * kasprintf(gfp_t gfp, const char *fmt, ...) { va_list ap; char *p; va_start(ap, fmt); p = kvasprintf(gfp, fmt, ap); va_end(ap); return (p); } static void linux_timer_callback_wrapper(void *context) { struct timer_list *timer; linux_set_current(curthread); timer = context; timer->function(timer->data); } int mod_timer(struct timer_list *timer, int expires) { int ret; timer->expires = expires; ret = callout_reset(&timer->callout, linux_timer_jiffies_until(expires), &linux_timer_callback_wrapper, timer); MPASS(ret == 0 || ret == 1); return (ret == 1); } void add_timer(struct timer_list *timer) { callout_reset(&timer->callout, linux_timer_jiffies_until(timer->expires), &linux_timer_callback_wrapper, timer); } void add_timer_on(struct timer_list *timer, int cpu) { callout_reset_on(&timer->callout, linux_timer_jiffies_until(timer->expires), &linux_timer_callback_wrapper, timer, cpu); } int del_timer(struct timer_list *timer) { if (callout_stop(&(timer)->callout) == -1) return (0); return (1); } int del_timer_sync(struct timer_list *timer) { if (callout_drain(&(timer)->callout) == -1) return (0); return (1); } /* greatest common divisor, Euclid equation */ static uint64_t lkpi_gcd_64(uint64_t a, uint64_t b) { uint64_t an; uint64_t bn; while (b != 0) { an = b; bn = a % b; a = an; b = bn; } return (a); } uint64_t lkpi_nsec2hz_rem; uint64_t lkpi_nsec2hz_div = 1000000000ULL; uint64_t lkpi_nsec2hz_max; uint64_t lkpi_usec2hz_rem; uint64_t lkpi_usec2hz_div = 1000000ULL; uint64_t lkpi_usec2hz_max; uint64_t lkpi_msec2hz_rem; uint64_t lkpi_msec2hz_div = 1000ULL; uint64_t lkpi_msec2hz_max; static void linux_timer_init(void *arg) { uint64_t gcd; /* * Compute an internal HZ value which can divide 2**32 to * avoid timer rounding problems when the tick value wraps * around 2**32: */ linux_timer_hz_mask = 1; while (linux_timer_hz_mask < (unsigned long)hz) linux_timer_hz_mask *= 2; linux_timer_hz_mask--; /* compute some internal constants */ lkpi_nsec2hz_rem = hz; lkpi_usec2hz_rem = hz; lkpi_msec2hz_rem = hz; gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div); lkpi_nsec2hz_rem /= gcd; lkpi_nsec2hz_div /= gcd; lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem; gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div); lkpi_usec2hz_rem /= gcd; lkpi_usec2hz_div /= gcd; lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem; gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div); lkpi_msec2hz_rem /= gcd; lkpi_msec2hz_div /= gcd; lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem; } SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL); void linux_complete_common(struct completion *c, int all) { int wakeup_swapper; sleepq_lock(c); if (all) { c->done = UINT_MAX; wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); } else { if (c->done != UINT_MAX) c->done++; wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); } sleepq_release(c); if (wakeup_swapper) kick_proc0(); } /* * Indefinite wait for done != 0 with or without signals. */ int linux_wait_for_common(struct completion *c, int flags) { struct task_struct *task; int error; if (SCHEDULER_STOPPED()) return (0); task = current; if (flags != 0) flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; else flags = SLEEPQ_SLEEP; error = 0; for (;;) { sleepq_lock(c); if (c->done) break; sleepq_add(c, NULL, "completion", flags, 0); if (flags & SLEEPQ_INTERRUPTIBLE) { DROP_GIANT(); error = -sleepq_wait_sig(c, 0); PICKUP_GIANT(); if (error != 0) { linux_schedule_save_interrupt_value(task, error); error = -ERESTARTSYS; goto intr; } } else { DROP_GIANT(); sleepq_wait(c, 0); PICKUP_GIANT(); } } if (c->done != UINT_MAX) c->done--; sleepq_release(c); intr: return (error); } /* * Time limited wait for done != 0 with or without signals. */ int linux_wait_for_timeout_common(struct completion *c, int timeout, int flags) { struct task_struct *task; int end = jiffies + timeout; int error; if (SCHEDULER_STOPPED()) return (0); task = current; if (flags != 0) flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; else flags = SLEEPQ_SLEEP; for (;;) { sleepq_lock(c); if (c->done) break; sleepq_add(c, NULL, "completion", flags, 0); sleepq_set_timeout(c, linux_timer_jiffies_until(end)); DROP_GIANT(); if (flags & SLEEPQ_INTERRUPTIBLE) error = -sleepq_timedwait_sig(c, 0); else error = -sleepq_timedwait(c, 0); PICKUP_GIANT(); if (error != 0) { /* check for timeout */ if (error == -EWOULDBLOCK) { error = 0; /* timeout */ } else { /* signal happened */ linux_schedule_save_interrupt_value(task, error); error = -ERESTARTSYS; } goto done; } } if (c->done != UINT_MAX) c->done--; sleepq_release(c); /* return how many jiffies are left */ error = linux_timer_jiffies_until(end); done: return (error); } int linux_try_wait_for_completion(struct completion *c) { int isdone; sleepq_lock(c); isdone = (c->done != 0); if (c->done != 0 && c->done != UINT_MAX) c->done--; sleepq_release(c); return (isdone); } int linux_completion_done(struct completion *c) { int isdone; sleepq_lock(c); isdone = (c->done != 0); sleepq_release(c); return (isdone); } static void linux_cdev_deref(struct linux_cdev *ldev) { if (refcount_release(&ldev->refs)) kfree(ldev); } static void linux_cdev_release(struct kobject *kobj) { struct linux_cdev *cdev; struct kobject *parent; cdev = container_of(kobj, struct linux_cdev, kobj); parent = kobj->parent; linux_destroy_dev(cdev); linux_cdev_deref(cdev); kobject_put(parent); } static void linux_cdev_static_release(struct kobject *kobj) { struct linux_cdev *cdev; struct kobject *parent; cdev = container_of(kobj, struct linux_cdev, kobj); parent = kobj->parent; linux_destroy_dev(cdev); kobject_put(parent); } void linux_destroy_dev(struct linux_cdev *ldev) { if (ldev->cdev == NULL) return; MPASS((ldev->siref & LDEV_SI_DTR) == 0); atomic_set_int(&ldev->siref, LDEV_SI_DTR); while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0) pause("ldevdtr", hz / 4); destroy_dev(ldev->cdev); ldev->cdev = NULL; } const struct kobj_type linux_cdev_ktype = { .release = linux_cdev_release, }; const struct kobj_type linux_cdev_static_ktype = { .release = linux_cdev_static_release, }; static void linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate) { struct notifier_block *nb; nb = arg; if (linkstate == LINK_STATE_UP) nb->notifier_call(nb, NETDEV_UP, ifp); else nb->notifier_call(nb, NETDEV_DOWN, ifp); } static void linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; nb = arg; nb->notifier_call(nb, NETDEV_REGISTER, ifp); } static void linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; nb = arg; nb->notifier_call(nb, NETDEV_UNREGISTER, ifp); } static void linux_handle_iflladdr_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; nb = arg; nb->notifier_call(nb, NETDEV_CHANGEADDR, ifp); } static void linux_handle_ifaddr_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; nb = arg; nb->notifier_call(nb, NETDEV_CHANGEIFADDR, ifp); } int register_netdevice_notifier(struct notifier_block *nb) { nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER( ifnet_link_event, linux_handle_ifnet_link_event, nb, 0); nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER( ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0); nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER( ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0); nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER( iflladdr_event, linux_handle_iflladdr_event, nb, 0); return (0); } int register_inetaddr_notifier(struct notifier_block *nb) { nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER( ifaddr_event, linux_handle_ifaddr_event, nb, 0); return (0); } int unregister_netdevice_notifier(struct notifier_block *nb) { EVENTHANDLER_DEREGISTER(ifnet_link_event, nb->tags[NETDEV_UP]); EVENTHANDLER_DEREGISTER(ifnet_arrival_event, nb->tags[NETDEV_REGISTER]); EVENTHANDLER_DEREGISTER(ifnet_departure_event, nb->tags[NETDEV_UNREGISTER]); EVENTHANDLER_DEREGISTER(iflladdr_event, nb->tags[NETDEV_CHANGEADDR]); return (0); } int unregister_inetaddr_notifier(struct notifier_block *nb) { EVENTHANDLER_DEREGISTER(ifaddr_event, nb->tags[NETDEV_CHANGEIFADDR]); return (0); } struct list_sort_thunk { int (*cmp)(void *, struct list_head *, struct list_head *); void *priv; }; static inline int linux_le_cmp(void *priv, const void *d1, const void *d2) { struct list_head *le1, *le2; struct list_sort_thunk *thunk; thunk = priv; le1 = *(__DECONST(struct list_head **, d1)); le2 = *(__DECONST(struct list_head **, d2)); return ((thunk->cmp)(thunk->priv, le1, le2)); } void list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv, struct list_head *a, struct list_head *b)) { struct list_sort_thunk thunk; struct list_head **ar, *le; size_t count, i; count = 0; list_for_each(le, head) count++; ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK); i = 0; list_for_each(le, head) ar[i++] = le; thunk.cmp = cmp; thunk.priv = priv; qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp); INIT_LIST_HEAD(head); for (i = 0; i < count; i++) list_add_tail(ar[i], head); free(ar, M_KMALLOC); } void linux_irq_handler(void *ent) { struct irq_ent *irqe; linux_set_current(curthread); irqe = ent; irqe->handler(irqe->irq, irqe->arg); } #if defined(__i386__) || defined(__amd64__) int linux_wbinvd_on_all_cpus(void) { pmap_invalidate_cache(); return (0); } #endif int linux_on_each_cpu(void callback(void *), void *data) { smp_rendezvous(smp_no_rendezvous_barrier, callback, smp_no_rendezvous_barrier, data); return (0); } int linux_in_atomic(void) { return ((curthread->td_pflags & TDP_NOFAULTING) != 0); } struct linux_cdev * linux_find_cdev(const char *name, unsigned major, unsigned minor) { dev_t dev = MKDEV(major, minor); struct cdev *cdev; dev_lock(); LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) { struct linux_cdev *ldev = cdev->si_drv1; if (ldev->dev == dev && strcmp(kobject_name(&ldev->kobj), name) == 0) { break; } } dev_unlock(); return (cdev != NULL ? cdev->si_drv1 : NULL); } int __register_chrdev(unsigned int major, unsigned int baseminor, unsigned int count, const char *name, const struct file_operations *fops) { struct linux_cdev *cdev; int ret = 0; int i; for (i = baseminor; i < baseminor + count; i++) { cdev = cdev_alloc(); cdev->ops = fops; kobject_set_name(&cdev->kobj, name); ret = cdev_add(cdev, makedev(major, i), 1); if (ret != 0) break; } return (ret); } int __register_chrdev_p(unsigned int major, unsigned int baseminor, unsigned int count, const char *name, const struct file_operations *fops, uid_t uid, gid_t gid, int mode) { struct linux_cdev *cdev; int ret = 0; int i; for (i = baseminor; i < baseminor + count; i++) { cdev = cdev_alloc(); cdev->ops = fops; kobject_set_name(&cdev->kobj, name); ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode); if (ret != 0) break; } return (ret); } void __unregister_chrdev(unsigned int major, unsigned int baseminor, unsigned int count, const char *name) { struct linux_cdev *cdevp; int i; for (i = baseminor; i < baseminor + count; i++) { cdevp = linux_find_cdev(name, major, i); if (cdevp != NULL) cdev_del(cdevp); } } void linux_dump_stack(void) { #ifdef STACK struct stack st; stack_zero(&st); stack_save(&st); stack_print(&st); #endif } #if defined(__i386__) || defined(__amd64__) bool linux_cpu_has_clflush; #endif static void linux_compat_init(void *arg) { struct sysctl_oid *rootoid; int i; #if defined(__i386__) || defined(__amd64__) linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH); #endif rw_init(&linux_vma_lock, "lkpi-vma-lock"); rootoid = SYSCTL_ADD_ROOT_NODE(NULL, OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys"); kobject_init(&linux_class_root, &linux_class_ktype); kobject_set_name(&linux_class_root, "class"); linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid), OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class"); kobject_init(&linux_root_device.kobj, &linux_dev_ktype); kobject_set_name(&linux_root_device.kobj, "device"); linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device"); linux_root_device.bsddev = root_bus; linux_class_misc.name = "misc"; class_register(&linux_class_misc); INIT_LIST_HEAD(&pci_drivers); INIT_LIST_HEAD(&pci_devices); spin_lock_init(&pci_lock); mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF); for (i = 0; i < VMMAP_HASH_SIZE; i++) LIST_INIT(&vmmaphead[i]); init_waitqueue_head(&linux_bit_waitq); init_waitqueue_head(&linux_var_waitq); } SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL); static void linux_compat_uninit(void *arg) { linux_kobject_kfree_name(&linux_class_root); linux_kobject_kfree_name(&linux_root_device.kobj); linux_kobject_kfree_name(&linux_class_misc.kobj); mtx_destroy(&vmmaplock); spin_lock_destroy(&pci_lock); rw_destroy(&linux_vma_lock); } SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL); /* * NOTE: Linux frequently uses "unsigned long" for pointer to integer * conversion and vice versa, where in FreeBSD "uintptr_t" would be * used. Assert these types have the same size, else some parts of the * LinuxKPI may not work like expected: */ CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t)); diff --git a/sys/dev/evdev/uinput.c b/sys/dev/evdev/uinput.c index 2449e80645f4..3edf14284a1f 100644 --- a/sys/dev/evdev/uinput.c +++ b/sys/dev/evdev/uinput.c @@ -1,721 +1,716 @@ /*- * Copyright (c) 2014 Jakub Wojciech Klama * Copyright (c) 2015-2016 Vladimir Kondratyev * 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. * * $FreeBSD$ */ #include "opt_evdev.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef UINPUT_DEBUG #define debugf(state, fmt, args...) printf("uinput: " fmt "\n", ##args) #else #define debugf(state, fmt, args...) #endif #define UINPUT_BUFFER_SIZE 16 #define UINPUT_LOCK(state) sx_xlock(&(state)->ucs_lock) #define UINPUT_UNLOCK(state) sx_unlock(&(state)->ucs_lock) #define UINPUT_LOCK_ASSERT(state) sx_assert(&(state)->ucs_lock, SA_LOCKED) #define UINPUT_EMPTYQ(state) \ ((state)->ucs_buffer_head == (state)->ucs_buffer_tail) enum uinput_state { UINPUT_NEW = 0, UINPUT_CONFIGURED, UINPUT_RUNNING }; static evdev_event_t uinput_ev_event; static d_open_t uinput_open; static d_read_t uinput_read; static d_write_t uinput_write; static d_ioctl_t uinput_ioctl; static d_poll_t uinput_poll; static d_kqfilter_t uinput_kqfilter; static void uinput_dtor(void *); static int uinput_kqread(struct knote *kn, long hint); static void uinput_kqdetach(struct knote *kn); static struct cdevsw uinput_cdevsw = { .d_version = D_VERSION, .d_open = uinput_open, .d_read = uinput_read, .d_write = uinput_write, .d_ioctl = uinput_ioctl, .d_poll = uinput_poll, .d_kqfilter = uinput_kqfilter, .d_name = "uinput", }; static struct cdev *uinput_cdev; static struct evdev_methods uinput_ev_methods = { .ev_open = NULL, .ev_close = NULL, .ev_event = uinput_ev_event, }; static struct filterops uinput_filterops = { .f_isfd = 1, .f_attach = NULL, .f_detach = uinput_kqdetach, .f_event = uinput_kqread, }; struct uinput_cdev_state { enum uinput_state ucs_state; struct evdev_dev * ucs_evdev; struct sx ucs_lock; size_t ucs_buffer_head; size_t ucs_buffer_tail; struct selinfo ucs_selp; bool ucs_blocked; bool ucs_selected; struct input_event ucs_buffer[UINPUT_BUFFER_SIZE]; }; static void uinput_enqueue_event(struct uinput_cdev_state *, uint16_t, uint16_t, int32_t); static int uinput_setup_provider(struct uinput_cdev_state *, struct uinput_user_dev *); static int uinput_cdev_create(void); static void uinput_notify(struct uinput_cdev_state *); static void uinput_knllock(void *arg) { struct sx *sx = arg; sx_xlock(sx); } static void uinput_knlunlock(void *arg) { struct sx *sx = arg; sx_unlock(sx); } static void -uinput_knl_assert_locked(void *arg) +uinput_knl_assert_lock(void *arg, int what) { - sx_assert((struct sx*)arg, SA_XLOCKED); -} - -static void -uinput_knl_assert_unlocked(void *arg) -{ - - sx_assert((struct sx*)arg, SA_UNLOCKED); + if (what == LA_LOCKED) + sx_assert((struct sx*)arg, SA_XLOCKED); + else + sx_assert((struct sx*)arg, SA_UNLOCKED); } static void uinput_ev_event(struct evdev_dev *evdev, uint16_t type, uint16_t code, int32_t value) { struct uinput_cdev_state *state = evdev_get_softc(evdev); if (type == EV_LED) evdev_push_event(evdev, type, code, value); UINPUT_LOCK(state); if (state->ucs_state == UINPUT_RUNNING) { uinput_enqueue_event(state, type, code, value); uinput_notify(state); } UINPUT_UNLOCK(state); } static void uinput_enqueue_event(struct uinput_cdev_state *state, uint16_t type, uint16_t code, int32_t value) { size_t head, tail; UINPUT_LOCK_ASSERT(state); head = state->ucs_buffer_head; tail = (state->ucs_buffer_tail + 1) % UINPUT_BUFFER_SIZE; microtime(&state->ucs_buffer[tail].time); state->ucs_buffer[tail].type = type; state->ucs_buffer[tail].code = code; state->ucs_buffer[tail].value = value; state->ucs_buffer_tail = tail; /* If queue is full remove oldest event */ if (tail == head) { debugf(state, "state %p: buffer overflow", state); head = (head + 1) % UINPUT_BUFFER_SIZE; state->ucs_buffer_head = head; } } static int uinput_open(struct cdev *dev, int oflags, int devtype, struct thread *td) { struct uinput_cdev_state *state; state = malloc(sizeof(struct uinput_cdev_state), M_EVDEV, M_WAITOK | M_ZERO); state->ucs_evdev = evdev_alloc(); sx_init(&state->ucs_lock, "uinput"); knlist_init(&state->ucs_selp.si_note, &state->ucs_lock, uinput_knllock, - uinput_knlunlock, uinput_knl_assert_locked, - uinput_knl_assert_unlocked); + uinput_knlunlock, uinput_knl_assert_lock); devfs_set_cdevpriv(state, uinput_dtor); return (0); } static void uinput_dtor(void *data) { struct uinput_cdev_state *state = (struct uinput_cdev_state *)data; evdev_free(state->ucs_evdev); knlist_clear(&state->ucs_selp.si_note, 0); seldrain(&state->ucs_selp); knlist_destroy(&state->ucs_selp.si_note); sx_destroy(&state->ucs_lock); free(data, M_EVDEV); } static int uinput_read(struct cdev *dev, struct uio *uio, int ioflag) { struct uinput_cdev_state *state; struct input_event *event; int remaining, ret; ret = devfs_get_cdevpriv((void **)&state); if (ret != 0) return (ret); debugf(state, "read %zd bytes by thread %d", uio->uio_resid, uio->uio_td->td_tid); /* Zero-sized reads are allowed for error checking */ if (uio->uio_resid != 0 && uio->uio_resid < sizeof(struct input_event)) return (EINVAL); remaining = uio->uio_resid / sizeof(struct input_event); UINPUT_LOCK(state); if (state->ucs_state != UINPUT_RUNNING) ret = EINVAL; if (ret == 0 && UINPUT_EMPTYQ(state)) { if (ioflag & O_NONBLOCK) ret = EWOULDBLOCK; else { if (remaining != 0) { state->ucs_blocked = true; ret = sx_sleep(state, &state->ucs_lock, PCATCH, "uiread", 0); } } } while (ret == 0 && !UINPUT_EMPTYQ(state) && remaining > 0) { event = &state->ucs_buffer[state->ucs_buffer_head]; state->ucs_buffer_head = (state->ucs_buffer_head + 1) % UINPUT_BUFFER_SIZE; remaining--; ret = uiomove(event, sizeof(struct input_event), uio); } UINPUT_UNLOCK(state); return (ret); } static int uinput_write(struct cdev *dev, struct uio *uio, int ioflag) { struct uinput_cdev_state *state; struct uinput_user_dev userdev; struct input_event event; int ret = 0; ret = devfs_get_cdevpriv((void **)&state); if (ret != 0) return (ret); debugf(state, "write %zd bytes by thread %d", uio->uio_resid, uio->uio_td->td_tid); UINPUT_LOCK(state); if (state->ucs_state != UINPUT_RUNNING) { /* Process written struct uinput_user_dev */ if (uio->uio_resid != sizeof(struct uinput_user_dev)) { debugf(state, "write size not multiple of " "struct uinput_user_dev size"); ret = EINVAL; } else { ret = uiomove(&userdev, sizeof(struct uinput_user_dev), uio); if (ret == 0) uinput_setup_provider(state, &userdev); } } else { /* Process written event */ if (uio->uio_resid % sizeof(struct input_event) != 0) { debugf(state, "write size not multiple of " "struct input_event size"); ret = EINVAL; } while (ret == 0 && uio->uio_resid > 0) { uiomove(&event, sizeof(struct input_event), uio); ret = evdev_push_event(state->ucs_evdev, event.type, event.code, event.value); } } UINPUT_UNLOCK(state); return (ret); } static int uinput_setup_dev(struct uinput_cdev_state *state, struct input_id *id, char *name, uint32_t ff_effects_max) { if (name[0] == 0) return (EINVAL); evdev_set_name(state->ucs_evdev, name); evdev_set_id(state->ucs_evdev, id->bustype, id->vendor, id->product, id->version); state->ucs_state = UINPUT_CONFIGURED; return (0); } static int uinput_setup_provider(struct uinput_cdev_state *state, struct uinput_user_dev *udev) { struct input_absinfo absinfo; int i, ret; debugf(state, "setup_provider called, udev=%p", udev); ret = uinput_setup_dev(state, &udev->id, udev->name, udev->ff_effects_max); if (ret) return (ret); bzero(&absinfo, sizeof(struct input_absinfo)); for (i = 0; i < ABS_CNT; i++) { if (!bit_test(state->ucs_evdev->ev_abs_flags, i)) continue; absinfo.minimum = udev->absmin[i]; absinfo.maximum = udev->absmax[i]; absinfo.fuzz = udev->absfuzz[i]; absinfo.flat = udev->absflat[i]; evdev_set_absinfo(state->ucs_evdev, i, &absinfo); } return (0); } static int uinput_poll(struct cdev *dev, int events, struct thread *td) { struct uinput_cdev_state *state; int revents = 0; if (devfs_get_cdevpriv((void **)&state) != 0) return (POLLNVAL); debugf(state, "poll by thread %d", td->td_tid); /* Always allow write */ if (events & (POLLOUT | POLLWRNORM)) revents |= (events & (POLLOUT | POLLWRNORM)); if (events & (POLLIN | POLLRDNORM)) { UINPUT_LOCK(state); if (!UINPUT_EMPTYQ(state)) revents = events & (POLLIN | POLLRDNORM); else { state->ucs_selected = true; selrecord(td, &state->ucs_selp); } UINPUT_UNLOCK(state); } return (revents); } static int uinput_kqfilter(struct cdev *dev, struct knote *kn) { struct uinput_cdev_state *state; int ret; ret = devfs_get_cdevpriv((void **)&state); if (ret != 0) return (ret); switch(kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &uinput_filterops; break; default: return(EINVAL); } kn->kn_hook = (caddr_t)state; knlist_add(&state->ucs_selp.si_note, kn, 0); return (0); } static int uinput_kqread(struct knote *kn, long hint) { struct uinput_cdev_state *state; int ret; state = (struct uinput_cdev_state *)kn->kn_hook; UINPUT_LOCK_ASSERT(state); ret = !UINPUT_EMPTYQ(state); return (ret); } static void uinput_kqdetach(struct knote *kn) { struct uinput_cdev_state *state; state = (struct uinput_cdev_state *)kn->kn_hook; knlist_remove(&state->ucs_selp.si_note, kn, 0); } static void uinput_notify(struct uinput_cdev_state *state) { UINPUT_LOCK_ASSERT(state); if (state->ucs_blocked) { state->ucs_blocked = false; wakeup(state); } if (state->ucs_selected) { state->ucs_selected = false; selwakeup(&state->ucs_selp); } KNOTE_LOCKED(&state->ucs_selp.si_note, 0); } static int uinput_ioctl_sub(struct uinput_cdev_state *state, u_long cmd, caddr_t data) { struct uinput_setup *us; struct uinput_abs_setup *uabs; int ret, len, intdata; char buf[NAMELEN]; UINPUT_LOCK_ASSERT(state); len = IOCPARM_LEN(cmd); if ((cmd & IOC_DIRMASK) == IOC_VOID && len == sizeof(int)) intdata = *(int *)data; switch (IOCBASECMD(cmd)) { case UI_GET_SYSNAME(0): if (state->ucs_state != UINPUT_RUNNING) return (ENOENT); if (len == 0) return (EINVAL); snprintf(data, len, "event%d", state->ucs_evdev->ev_unit); return (0); } switch (cmd) { case UI_DEV_CREATE: if (state->ucs_state != UINPUT_CONFIGURED) return (EINVAL); evdev_set_methods(state->ucs_evdev, state, &uinput_ev_methods); evdev_set_flag(state->ucs_evdev, EVDEV_FLAG_SOFTREPEAT); ret = evdev_register(state->ucs_evdev); if (ret == 0) state->ucs_state = UINPUT_RUNNING; return (ret); case UI_DEV_DESTROY: if (state->ucs_state != UINPUT_RUNNING) return (0); evdev_unregister(state->ucs_evdev); bzero(state->ucs_evdev, sizeof(struct evdev_dev)); state->ucs_state = UINPUT_NEW; return (0); case UI_DEV_SETUP: if (state->ucs_state == UINPUT_RUNNING) return (EINVAL); us = (struct uinput_setup *)data; return (uinput_setup_dev(state, &us->id, us->name, us->ff_effects_max)); case UI_ABS_SETUP: if (state->ucs_state == UINPUT_RUNNING) return (EINVAL); uabs = (struct uinput_abs_setup *)data; if (uabs->code > ABS_MAX) return (EINVAL); evdev_support_abs(state->ucs_evdev, uabs->code, uabs->absinfo.value, uabs->absinfo.minimum, uabs->absinfo.maximum, uabs->absinfo.fuzz, uabs->absinfo.flat, uabs->absinfo.resolution); return (0); case UI_SET_EVBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > EV_MAX || intdata < 0) return (EINVAL); evdev_support_event(state->ucs_evdev, intdata); return (0); case UI_SET_KEYBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > KEY_MAX || intdata < 0) return (EINVAL); evdev_support_key(state->ucs_evdev, intdata); return (0); case UI_SET_RELBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > REL_MAX || intdata < 0) return (EINVAL); evdev_support_rel(state->ucs_evdev, intdata); return (0); case UI_SET_ABSBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > ABS_MAX || intdata < 0) return (EINVAL); evdev_set_abs_bit(state->ucs_evdev, intdata); return (0); case UI_SET_MSCBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > MSC_MAX || intdata < 0) return (EINVAL); evdev_support_msc(state->ucs_evdev, intdata); return (0); case UI_SET_LEDBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > LED_MAX || intdata < 0) return (EINVAL); evdev_support_led(state->ucs_evdev, intdata); return (0); case UI_SET_SNDBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > SND_MAX || intdata < 0) return (EINVAL); evdev_support_snd(state->ucs_evdev, intdata); return (0); case UI_SET_FFBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > FF_MAX || intdata < 0) return (EINVAL); /* Fake unsupported ioctl */ return (0); case UI_SET_PHYS: if (state->ucs_state == UINPUT_RUNNING) return (EINVAL); ret = copyinstr(*(void **)data, buf, sizeof(buf), NULL); /* Linux returns EINVAL when string does not fit the buffer */ if (ret == ENAMETOOLONG) ret = EINVAL; if (ret != 0) return (ret); evdev_set_phys(state->ucs_evdev, buf); return (0); case UI_SET_BSDUNIQ: if (state->ucs_state == UINPUT_RUNNING) return (EINVAL); ret = copyinstr(*(void **)data, buf, sizeof(buf), NULL); if (ret != 0) return (ret); evdev_set_serial(state->ucs_evdev, buf); return (0); case UI_SET_SWBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > SW_MAX || intdata < 0) return (EINVAL); evdev_support_sw(state->ucs_evdev, intdata); return (0); case UI_SET_PROPBIT: if (state->ucs_state == UINPUT_RUNNING || intdata > INPUT_PROP_MAX || intdata < 0) return (EINVAL); evdev_support_prop(state->ucs_evdev, intdata); return (0); case UI_BEGIN_FF_UPLOAD: case UI_END_FF_UPLOAD: case UI_BEGIN_FF_ERASE: case UI_END_FF_ERASE: if (state->ucs_state == UINPUT_RUNNING) return (EINVAL); /* Fake unsupported ioctl */ return (0); case UI_GET_VERSION: *(unsigned int *)data = UINPUT_VERSION; return (0); } return (EINVAL); } static int uinput_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) { struct uinput_cdev_state *state; int ret; ret = devfs_get_cdevpriv((void **)&state); if (ret != 0) return (ret); debugf(state, "ioctl called: cmd=0x%08lx, data=%p", cmd, data); UINPUT_LOCK(state); ret = uinput_ioctl_sub(state, cmd, data); UINPUT_UNLOCK(state); return (ret); } static int uinput_cdev_create(void) { struct make_dev_args mda; int ret; make_dev_args_init(&mda); mda.mda_flags = MAKEDEV_WAITOK | MAKEDEV_CHECKNAME; mda.mda_devsw = &uinput_cdevsw; mda.mda_uid = UID_ROOT; mda.mda_gid = GID_WHEEL; mda.mda_mode = 0600; ret = make_dev_s(&mda, &uinput_cdev, "uinput"); return (ret); } static int uinput_cdev_destroy(void) { destroy_dev(uinput_cdev); return (0); } static int uinput_modevent(module_t mod __unused, int cmd, void *data) { int ret = 0; switch (cmd) { case MOD_LOAD: ret = uinput_cdev_create(); break; case MOD_UNLOAD: ret = uinput_cdev_destroy(); break; case MOD_SHUTDOWN: break; default: ret = EINVAL; break; } return (ret); } DEV_MODULE(uinput, uinput_modevent, NULL); MODULE_VERSION(uinput, 1); MODULE_DEPEND(uinput, evdev, 1, 1, 1); diff --git a/sys/kern/kern_event.c b/sys/kern/kern_event.c index 5c121b1f4e81..a1a4a3c470b8 100644 --- a/sys/kern/kern_event.c +++ b/sys/kern/kern_event.c @@ -1,2741 +1,2729 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1999,2000,2001 Jonathan Lemon * Copyright 2004 John-Mark Gurney * Copyright (c) 2009 Apple, Inc. * 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_ktrace.h" #include "opt_kqueue.h" #ifdef COMPAT_FREEBSD11 #define _WANT_FREEBSD11_KEVENT #endif #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 KTRACE #include #endif #include #include static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); /* * This lock is used if multiple kq locks are required. This possibly * should be made into a per proc lock. */ static struct mtx kq_global; MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); #define KQ_GLOBAL_LOCK(lck, haslck) do { \ if (!haslck) \ mtx_lock(lck); \ haslck = 1; \ } while (0) #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ if (haslck) \ mtx_unlock(lck); \ haslck = 0; \ } while (0) TASKQUEUE_DEFINE_THREAD(kqueue_ctx); static int kevent_copyout(void *arg, struct kevent *kevp, int count); static int kevent_copyin(void *arg, struct kevent *kevp, int count); static int kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int mflag); static int kqueue_acquire(struct file *fp, struct kqueue **kqp); static void kqueue_release(struct kqueue *kq, int locked); static void kqueue_destroy(struct kqueue *kq); static void kqueue_drain(struct kqueue *kq, struct thread *td); static int kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, int mflag); static void kqueue_task(void *arg, int pending); static int kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, const struct timespec *timeout, struct kevent *keva, struct thread *td); static void kqueue_wakeup(struct kqueue *kq); static struct filterops *kqueue_fo_find(int filt); static void kqueue_fo_release(int filt); struct g_kevent_args; static int kern_kevent_generic(struct thread *td, struct g_kevent_args *uap, struct kevent_copyops *k_ops, const char *struct_name); static fo_ioctl_t kqueue_ioctl; static fo_poll_t kqueue_poll; static fo_kqfilter_t kqueue_kqfilter; static fo_stat_t kqueue_stat; static fo_close_t kqueue_close; static fo_fill_kinfo_t kqueue_fill_kinfo; static struct fileops kqueueops = { .fo_read = invfo_rdwr, .fo_write = invfo_rdwr, .fo_truncate = invfo_truncate, .fo_ioctl = kqueue_ioctl, .fo_poll = kqueue_poll, .fo_kqfilter = kqueue_kqfilter, .fo_stat = kqueue_stat, .fo_close = kqueue_close, .fo_chmod = invfo_chmod, .fo_chown = invfo_chown, .fo_sendfile = invfo_sendfile, .fo_fill_kinfo = kqueue_fill_kinfo, }; static int knote_attach(struct knote *kn, struct kqueue *kq); static void knote_drop(struct knote *kn, struct thread *td); static void knote_drop_detached(struct knote *kn, struct thread *td); static void knote_enqueue(struct knote *kn); static void knote_dequeue(struct knote *kn); static void knote_init(void); static struct knote *knote_alloc(int mflag); static void knote_free(struct knote *kn); static void filt_kqdetach(struct knote *kn); static int filt_kqueue(struct knote *kn, long hint); static int filt_procattach(struct knote *kn); static void filt_procdetach(struct knote *kn); static int filt_proc(struct knote *kn, long hint); static int filt_fileattach(struct knote *kn); static void filt_timerexpire(void *knx); static int filt_timerattach(struct knote *kn); static void filt_timerdetach(struct knote *kn); static void filt_timerstart(struct knote *kn, sbintime_t to); static void filt_timertouch(struct knote *kn, struct kevent *kev, u_long type); static int filt_timervalidate(struct knote *kn, sbintime_t *to); static int filt_timer(struct knote *kn, long hint); static int filt_userattach(struct knote *kn); static void filt_userdetach(struct knote *kn); static int filt_user(struct knote *kn, long hint); static void filt_usertouch(struct knote *kn, struct kevent *kev, u_long type); static struct filterops file_filtops = { .f_isfd = 1, .f_attach = filt_fileattach, }; static struct filterops kqread_filtops = { .f_isfd = 1, .f_detach = filt_kqdetach, .f_event = filt_kqueue, }; /* XXX - move to kern_proc.c? */ static struct filterops proc_filtops = { .f_isfd = 0, .f_attach = filt_procattach, .f_detach = filt_procdetach, .f_event = filt_proc, }; static struct filterops timer_filtops = { .f_isfd = 0, .f_attach = filt_timerattach, .f_detach = filt_timerdetach, .f_event = filt_timer, .f_touch = filt_timertouch, }; static struct filterops user_filtops = { .f_attach = filt_userattach, .f_detach = filt_userdetach, .f_event = filt_user, .f_touch = filt_usertouch, }; static uma_zone_t knote_zone; static unsigned int kq_ncallouts = 0; static unsigned int kq_calloutmax = 4 * 1024; SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); /* XXX - ensure not influx ? */ #define KNOTE_ACTIVATE(kn, islock) do { \ if ((islock)) \ mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ else \ KQ_LOCK((kn)->kn_kq); \ (kn)->kn_status |= KN_ACTIVE; \ if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ knote_enqueue((kn)); \ if (!(islock)) \ KQ_UNLOCK((kn)->kn_kq); \ } while(0) #define KQ_LOCK(kq) do { \ mtx_lock(&(kq)->kq_lock); \ } while (0) #define KQ_FLUX_WAKEUP(kq) do { \ if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ (kq)->kq_state &= ~KQ_FLUXWAIT; \ wakeup((kq)); \ } \ } while (0) #define KQ_UNLOCK_FLUX(kq) do { \ KQ_FLUX_WAKEUP(kq); \ mtx_unlock(&(kq)->kq_lock); \ } while (0) #define KQ_UNLOCK(kq) do { \ mtx_unlock(&(kq)->kq_lock); \ } while (0) #define KQ_OWNED(kq) do { \ mtx_assert(&(kq)->kq_lock, MA_OWNED); \ } while (0) #define KQ_NOTOWNED(kq) do { \ mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ } while (0) static struct knlist * kn_list_lock(struct knote *kn) { struct knlist *knl; knl = kn->kn_knlist; if (knl != NULL) knl->kl_lock(knl->kl_lockarg); return (knl); } static void kn_list_unlock(struct knlist *knl) { bool do_free; if (knl == NULL) return; do_free = knl->kl_autodestroy && knlist_empty(knl); knl->kl_unlock(knl->kl_lockarg); if (do_free) { knlist_destroy(knl); free(knl, M_KQUEUE); } } static bool kn_in_flux(struct knote *kn) { return (kn->kn_influx > 0); } static void kn_enter_flux(struct knote *kn) { KQ_OWNED(kn->kn_kq); MPASS(kn->kn_influx < INT_MAX); kn->kn_influx++; } static bool kn_leave_flux(struct knote *kn) { KQ_OWNED(kn->kn_kq); MPASS(kn->kn_influx > 0); kn->kn_influx--; return (kn->kn_influx == 0); } #define KNL_ASSERT_LOCK(knl, islocked) do { \ if (islocked) \ KNL_ASSERT_LOCKED(knl); \ else \ KNL_ASSERT_UNLOCKED(knl); \ } while (0) #ifdef INVARIANTS #define KNL_ASSERT_LOCKED(knl) do { \ - knl->kl_assert_locked((knl)->kl_lockarg); \ + knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED); \ } while (0) #define KNL_ASSERT_UNLOCKED(knl) do { \ - knl->kl_assert_unlocked((knl)->kl_lockarg); \ + knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED); \ } while (0) #else /* !INVARIANTS */ #define KNL_ASSERT_LOCKED(knl) do {} while(0) #define KNL_ASSERT_UNLOCKED(knl) do {} while (0) #endif /* INVARIANTS */ #ifndef KN_HASHSIZE #define KN_HASHSIZE 64 /* XXX should be tunable */ #endif #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) static int filt_nullattach(struct knote *kn) { return (ENXIO); }; struct filterops null_filtops = { .f_isfd = 0, .f_attach = filt_nullattach, }; /* XXX - make SYSINIT to add these, and move into respective modules. */ extern struct filterops sig_filtops; extern struct filterops fs_filtops; /* * Table for for all system-defined filters. */ static struct mtx filterops_lock; MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF); static struct { struct filterops *for_fop; int for_nolock; int for_refcnt; } sysfilt_ops[EVFILT_SYSCOUNT] = { { &file_filtops, 1 }, /* EVFILT_READ */ { &file_filtops, 1 }, /* EVFILT_WRITE */ { &null_filtops }, /* EVFILT_AIO */ { &file_filtops, 1 }, /* EVFILT_VNODE */ { &proc_filtops, 1 }, /* EVFILT_PROC */ { &sig_filtops, 1 }, /* EVFILT_SIGNAL */ { &timer_filtops, 1 }, /* EVFILT_TIMER */ { &file_filtops, 1 }, /* EVFILT_PROCDESC */ { &fs_filtops, 1 }, /* EVFILT_FS */ { &null_filtops }, /* EVFILT_LIO */ { &user_filtops, 1 }, /* EVFILT_USER */ { &null_filtops }, /* EVFILT_SENDFILE */ { &file_filtops, 1 }, /* EVFILT_EMPTY */ }; /* * Simple redirection for all cdevsw style objects to call their fo_kqfilter * method. */ static int filt_fileattach(struct knote *kn) { return (fo_kqfilter(kn->kn_fp, kn)); } /*ARGSUSED*/ static int kqueue_kqfilter(struct file *fp, struct knote *kn) { struct kqueue *kq = kn->kn_fp->f_data; if (kn->kn_filter != EVFILT_READ) return (EINVAL); kn->kn_status |= KN_KQUEUE; kn->kn_fop = &kqread_filtops; knlist_add(&kq->kq_sel.si_note, kn, 0); return (0); } static void filt_kqdetach(struct knote *kn) { struct kqueue *kq = kn->kn_fp->f_data; knlist_remove(&kq->kq_sel.si_note, kn, 0); } /*ARGSUSED*/ static int filt_kqueue(struct knote *kn, long hint) { struct kqueue *kq = kn->kn_fp->f_data; kn->kn_data = kq->kq_count; return (kn->kn_data > 0); } /* XXX - move to kern_proc.c? */ static int filt_procattach(struct knote *kn) { struct proc *p; int error; bool exiting, immediate; exiting = immediate = false; if (kn->kn_sfflags & NOTE_EXIT) p = pfind_any(kn->kn_id); else p = pfind(kn->kn_id); if (p == NULL) return (ESRCH); if (p->p_flag & P_WEXIT) exiting = true; if ((error = p_cansee(curthread, p))) { PROC_UNLOCK(p); return (error); } kn->kn_ptr.p_proc = p; kn->kn_flags |= EV_CLEAR; /* automatically set */ /* * Internal flag indicating registration done by kernel for the * purposes of getting a NOTE_CHILD notification. */ if (kn->kn_flags & EV_FLAG2) { kn->kn_flags &= ~EV_FLAG2; kn->kn_data = kn->kn_sdata; /* ppid */ kn->kn_fflags = NOTE_CHILD; kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK); immediate = true; /* Force immediate activation of child note. */ } /* * Internal flag indicating registration done by kernel (for other than * NOTE_CHILD). */ if (kn->kn_flags & EV_FLAG1) { kn->kn_flags &= ~EV_FLAG1; } knlist_add(p->p_klist, kn, 1); /* * Immediately activate any child notes or, in the case of a zombie * target process, exit notes. The latter is necessary to handle the * case where the target process, e.g. a child, dies before the kevent * is registered. */ if (immediate || (exiting && filt_proc(kn, NOTE_EXIT))) KNOTE_ACTIVATE(kn, 0); PROC_UNLOCK(p); return (0); } /* * The knote may be attached to a different process, which may exit, * leaving nothing for the knote to be attached to. So when the process * exits, the knote is marked as DETACHED and also flagged as ONESHOT so * it will be deleted when read out. However, as part of the knote deletion, * this routine is called, so a check is needed to avoid actually performing * a detach, because the original process does not exist any more. */ /* XXX - move to kern_proc.c? */ static void filt_procdetach(struct knote *kn) { knlist_remove(kn->kn_knlist, kn, 0); kn->kn_ptr.p_proc = NULL; } /* XXX - move to kern_proc.c? */ static int filt_proc(struct knote *kn, long hint) { struct proc *p; u_int event; p = kn->kn_ptr.p_proc; if (p == NULL) /* already activated, from attach filter */ return (0); /* Mask off extra data. */ event = (u_int)hint & NOTE_PCTRLMASK; /* If the user is interested in this event, record it. */ if (kn->kn_sfflags & event) kn->kn_fflags |= event; /* Process is gone, so flag the event as finished. */ if (event == NOTE_EXIT) { kn->kn_flags |= EV_EOF | EV_ONESHOT; kn->kn_ptr.p_proc = NULL; if (kn->kn_fflags & NOTE_EXIT) kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig); if (kn->kn_fflags == 0) kn->kn_flags |= EV_DROP; return (1); } return (kn->kn_fflags != 0); } /* * Called when the process forked. It mostly does the same as the * knote(), activating all knotes registered to be activated when the * process forked. Additionally, for each knote attached to the * parent, check whether user wants to track the new process. If so * attach a new knote to it, and immediately report an event with the * child's pid. */ void knote_fork(struct knlist *list, int pid) { struct kqueue *kq; struct knote *kn; struct kevent kev; int error; MPASS(list != NULL); KNL_ASSERT_LOCKED(list); if (SLIST_EMPTY(&list->kl_list)) return; memset(&kev, 0, sizeof(kev)); SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { kq = kn->kn_kq; KQ_LOCK(kq); if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { KQ_UNLOCK(kq); continue; } /* * The same as knote(), activate the event. */ if ((kn->kn_sfflags & NOTE_TRACK) == 0) { if (kn->kn_fop->f_event(kn, NOTE_FORK)) KNOTE_ACTIVATE(kn, 1); KQ_UNLOCK(kq); continue; } /* * The NOTE_TRACK case. In addition to the activation * of the event, we need to register new events to * track the child. Drop the locks in preparation for * the call to kqueue_register(). */ kn_enter_flux(kn); KQ_UNLOCK(kq); list->kl_unlock(list->kl_lockarg); /* * Activate existing knote and register tracking knotes with * new process. * * First register a knote to get just the child notice. This * must be a separate note from a potential NOTE_EXIT * notification since both NOTE_CHILD and NOTE_EXIT are defined * to use the data field (in conflicting ways). */ kev.ident = pid; kev.filter = kn->kn_filter; kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT | EV_FLAG2; kev.fflags = kn->kn_sfflags; kev.data = kn->kn_id; /* parent */ kev.udata = kn->kn_kevent.udata;/* preserve udata */ error = kqueue_register(kq, &kev, NULL, M_NOWAIT); if (error) kn->kn_fflags |= NOTE_TRACKERR; /* * Then register another knote to track other potential events * from the new process. */ kev.ident = pid; kev.filter = kn->kn_filter; kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; kev.fflags = kn->kn_sfflags; kev.data = kn->kn_id; /* parent */ kev.udata = kn->kn_kevent.udata;/* preserve udata */ error = kqueue_register(kq, &kev, NULL, M_NOWAIT); if (error) kn->kn_fflags |= NOTE_TRACKERR; if (kn->kn_fop->f_event(kn, NOTE_FORK)) KNOTE_ACTIVATE(kn, 0); list->kl_lock(list->kl_lockarg); KQ_LOCK(kq); kn_leave_flux(kn); KQ_UNLOCK_FLUX(kq); } } /* * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the * interval timer support code. */ #define NOTE_TIMER_PRECMASK \ (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS) static sbintime_t timer2sbintime(int64_t data, int flags) { int64_t secs; /* * Macros for converting to the fractional second portion of an * sbintime_t using 64bit multiplication to improve precision. */ #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32) #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32) #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32) switch (flags & NOTE_TIMER_PRECMASK) { case NOTE_SECONDS: #ifdef __LP64__ if (data > (SBT_MAX / SBT_1S)) return (SBT_MAX); #endif return ((sbintime_t)data << 32); case NOTE_MSECONDS: /* FALLTHROUGH */ case 0: if (data >= 1000) { secs = data / 1000; #ifdef __LP64__ if (secs > (SBT_MAX / SBT_1S)) return (SBT_MAX); #endif return (secs << 32 | MS_TO_SBT(data % 1000)); } return (MS_TO_SBT(data)); case NOTE_USECONDS: if (data >= 1000000) { secs = data / 1000000; #ifdef __LP64__ if (secs > (SBT_MAX / SBT_1S)) return (SBT_MAX); #endif return (secs << 32 | US_TO_SBT(data % 1000000)); } return (US_TO_SBT(data)); case NOTE_NSECONDS: if (data >= 1000000000) { secs = data / 1000000000; #ifdef __LP64__ if (secs > (SBT_MAX / SBT_1S)) return (SBT_MAX); #endif return (secs << 32 | US_TO_SBT(data % 1000000000)); } return (NS_TO_SBT(data)); default: break; } return (-1); } struct kq_timer_cb_data { struct callout c; sbintime_t next; /* next timer event fires at */ sbintime_t to; /* precalculated timer period, 0 for abs */ }; static void filt_timerexpire(void *knx) { struct knote *kn; struct kq_timer_cb_data *kc; kn = knx; kn->kn_data++; KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ if ((kn->kn_flags & EV_ONESHOT) != 0) return; kc = kn->kn_ptr.p_v; if (kc->to == 0) return; kc->next += kc->to; callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE); } /* * data contains amount of time to sleep */ static int filt_timervalidate(struct knote *kn, sbintime_t *to) { struct bintime bt; sbintime_t sbt; if (kn->kn_sdata < 0) return (EINVAL); if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) kn->kn_sdata = 1; /* * The only fflags values supported are the timer unit * (precision) and the absolute time indicator. */ if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0) return (EINVAL); *to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { getboottimebin(&bt); sbt = bttosbt(bt); *to -= sbt; } if (*to < 0) return (EINVAL); return (0); } static int filt_timerattach(struct knote *kn) { struct kq_timer_cb_data *kc; sbintime_t to; unsigned int ncallouts; int error; error = filt_timervalidate(kn, &to); if (error != 0) return (error); do { ncallouts = kq_ncallouts; if (ncallouts >= kq_calloutmax) return (ENOMEM); } while (!atomic_cmpset_int(&kq_ncallouts, ncallouts, ncallouts + 1)); if ((kn->kn_sfflags & NOTE_ABSTIME) == 0) kn->kn_flags |= EV_CLEAR; /* automatically set */ kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK); callout_init(&kc->c, 1); filt_timerstart(kn, to); return (0); } static void filt_timerstart(struct knote *kn, sbintime_t to) { struct kq_timer_cb_data *kc; kc = kn->kn_ptr.p_v; if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { kc->next = to; kc->to = 0; } else { kc->next = to + sbinuptime(); kc->to = to; } callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE); } static void filt_timerdetach(struct knote *kn) { struct kq_timer_cb_data *kc; unsigned int old __unused; kc = kn->kn_ptr.p_v; callout_drain(&kc->c); free(kc, M_KQUEUE); old = atomic_fetchadd_int(&kq_ncallouts, -1); KASSERT(old > 0, ("Number of callouts cannot become negative")); kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ } static void filt_timertouch(struct knote *kn, struct kevent *kev, u_long type) { struct kq_timer_cb_data *kc; struct kqueue *kq; sbintime_t to; int error; switch (type) { case EVENT_REGISTER: /* Handle re-added timers that update data/fflags */ if (kev->flags & EV_ADD) { kc = kn->kn_ptr.p_v; /* Drain any existing callout. */ callout_drain(&kc->c); /* Throw away any existing undelivered record * of the timer expiration. This is done under * the presumption that if a process is * re-adding this timer with new parameters, * it is no longer interested in what may have * happened under the old parameters. If it is * interested, it can wait for the expiration, * delete the old timer definition, and then * add the new one. * * This has to be done while the kq is locked: * - if enqueued, dequeue * - make it no longer active * - clear the count of expiration events */ kq = kn->kn_kq; KQ_LOCK(kq); if (kn->kn_status & KN_QUEUED) knote_dequeue(kn); kn->kn_status &= ~KN_ACTIVE; kn->kn_data = 0; KQ_UNLOCK(kq); /* Reschedule timer based on new data/fflags */ kn->kn_sfflags = kev->fflags; kn->kn_sdata = kev->data; error = filt_timervalidate(kn, &to); if (error != 0) { kn->kn_flags |= EV_ERROR; kn->kn_data = error; } else filt_timerstart(kn, to); } break; case EVENT_PROCESS: *kev = kn->kn_kevent; if (kn->kn_flags & EV_CLEAR) { kn->kn_data = 0; kn->kn_fflags = 0; } break; default: panic("filt_timertouch() - invalid type (%ld)", type); break; } } static int filt_timer(struct knote *kn, long hint) { return (kn->kn_data != 0); } static int filt_userattach(struct knote *kn) { /* * EVFILT_USER knotes are not attached to anything in the kernel. */ kn->kn_hook = NULL; if (kn->kn_fflags & NOTE_TRIGGER) kn->kn_hookid = 1; else kn->kn_hookid = 0; return (0); } static void filt_userdetach(__unused struct knote *kn) { /* * EVFILT_USER knotes are not attached to anything in the kernel. */ } static int filt_user(struct knote *kn, __unused long hint) { return (kn->kn_hookid); } static void filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) { u_int ffctrl; switch (type) { case EVENT_REGISTER: if (kev->fflags & NOTE_TRIGGER) kn->kn_hookid = 1; ffctrl = kev->fflags & NOTE_FFCTRLMASK; kev->fflags &= NOTE_FFLAGSMASK; switch (ffctrl) { case NOTE_FFNOP: break; case NOTE_FFAND: kn->kn_sfflags &= kev->fflags; break; case NOTE_FFOR: kn->kn_sfflags |= kev->fflags; break; case NOTE_FFCOPY: kn->kn_sfflags = kev->fflags; break; default: /* XXX Return error? */ break; } kn->kn_sdata = kev->data; if (kev->flags & EV_CLEAR) { kn->kn_hookid = 0; kn->kn_data = 0; kn->kn_fflags = 0; } break; case EVENT_PROCESS: *kev = kn->kn_kevent; kev->fflags = kn->kn_sfflags; kev->data = kn->kn_sdata; if (kn->kn_flags & EV_CLEAR) { kn->kn_hookid = 0; kn->kn_data = 0; kn->kn_fflags = 0; } break; default: panic("filt_usertouch() - invalid type (%ld)", type); break; } } int sys_kqueue(struct thread *td, struct kqueue_args *uap) { return (kern_kqueue(td, 0, NULL)); } static void kqueue_init(struct kqueue *kq) { mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK); TAILQ_INIT(&kq->kq_head); knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); } int kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps) { struct filedesc *fdp; struct kqueue *kq; struct file *fp; struct ucred *cred; int fd, error; fdp = td->td_proc->p_fd; cred = td->td_ucred; if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES))) return (ENOMEM); error = falloc_caps(td, &fp, &fd, flags, fcaps); if (error != 0) { chgkqcnt(cred->cr_ruidinfo, -1, 0); return (error); } /* An extra reference on `fp' has been held for us by falloc(). */ kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); kqueue_init(kq); kq->kq_fdp = fdp; kq->kq_cred = crhold(cred); FILEDESC_XLOCK(fdp); TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); FILEDESC_XUNLOCK(fdp); finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); fdrop(fp, td); td->td_retval[0] = fd; return (0); } struct g_kevent_args { int fd; void *changelist; int nchanges; void *eventlist; int nevents; const struct timespec *timeout; }; int sys_kevent(struct thread *td, struct kevent_args *uap) { struct kevent_copyops k_ops = { .arg = uap, .k_copyout = kevent_copyout, .k_copyin = kevent_copyin, .kevent_size = sizeof(struct kevent), }; struct g_kevent_args gk_args = { .fd = uap->fd, .changelist = uap->changelist, .nchanges = uap->nchanges, .eventlist = uap->eventlist, .nevents = uap->nevents, .timeout = uap->timeout, }; return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent")); } static int kern_kevent_generic(struct thread *td, struct g_kevent_args *uap, struct kevent_copyops *k_ops, const char *struct_name) { struct timespec ts, *tsp; #ifdef KTRACE struct kevent *eventlist = uap->eventlist; #endif int error; if (uap->timeout != NULL) { error = copyin(uap->timeout, &ts, sizeof(ts)); if (error) return (error); tsp = &ts; } else tsp = NULL; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT_ARRAY)) ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist, uap->nchanges, k_ops->kevent_size); #endif error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, k_ops, tsp); #ifdef KTRACE if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) ktrstructarray(struct_name, UIO_USERSPACE, eventlist, td->td_retval[0], k_ops->kevent_size); #endif return (error); } /* * Copy 'count' items into the destination list pointed to by uap->eventlist. */ static int kevent_copyout(void *arg, struct kevent *kevp, int count) { struct kevent_args *uap; int error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct kevent_args *)arg; error = copyout(kevp, uap->eventlist, count * sizeof *kevp); if (error == 0) uap->eventlist += count; return (error); } /* * Copy 'count' items from the list pointed to by uap->changelist. */ static int kevent_copyin(void *arg, struct kevent *kevp, int count) { struct kevent_args *uap; int error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct kevent_args *)arg; error = copyin(uap->changelist, kevp, count * sizeof *kevp); if (error == 0) uap->changelist += count; return (error); } #ifdef COMPAT_FREEBSD11 static int kevent11_copyout(void *arg, struct kevent *kevp, int count) { struct freebsd11_kevent_args *uap; struct kevent_freebsd11 kev11; int error, i; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct freebsd11_kevent_args *)arg; for (i = 0; i < count; i++) { kev11.ident = kevp->ident; kev11.filter = kevp->filter; kev11.flags = kevp->flags; kev11.fflags = kevp->fflags; kev11.data = kevp->data; kev11.udata = kevp->udata; error = copyout(&kev11, uap->eventlist, sizeof(kev11)); if (error != 0) break; uap->eventlist++; kevp++; } return (error); } /* * Copy 'count' items from the list pointed to by uap->changelist. */ static int kevent11_copyin(void *arg, struct kevent *kevp, int count) { struct freebsd11_kevent_args *uap; struct kevent_freebsd11 kev11; int error, i; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct freebsd11_kevent_args *)arg; for (i = 0; i < count; i++) { error = copyin(uap->changelist, &kev11, sizeof(kev11)); if (error != 0) break; kevp->ident = kev11.ident; kevp->filter = kev11.filter; kevp->flags = kev11.flags; kevp->fflags = kev11.fflags; kevp->data = (uintptr_t)kev11.data; kevp->udata = kev11.udata; bzero(&kevp->ext, sizeof(kevp->ext)); uap->changelist++; kevp++; } return (error); } int freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap) { struct kevent_copyops k_ops = { .arg = uap, .k_copyout = kevent11_copyout, .k_copyin = kevent11_copyin, .kevent_size = sizeof(struct kevent_freebsd11), }; struct g_kevent_args gk_args = { .fd = uap->fd, .changelist = uap->changelist, .nchanges = uap->nchanges, .eventlist = uap->eventlist, .nevents = uap->nevents, .timeout = uap->timeout, }; return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent_freebsd11")); } #endif int kern_kevent(struct thread *td, int fd, int nchanges, int nevents, struct kevent_copyops *k_ops, const struct timespec *timeout) { cap_rights_t rights; struct file *fp; int error; cap_rights_init_zero(&rights); if (nchanges > 0) cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE); if (nevents > 0) cap_rights_set_one(&rights, CAP_KQUEUE_EVENT); error = fget(td, fd, &rights, &fp); if (error != 0) return (error); error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout); fdrop(fp, td); return (error); } static int kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents, struct kevent_copyops *k_ops, const struct timespec *timeout) { struct kevent keva[KQ_NEVENTS]; struct kevent *kevp, *changes; int i, n, nerrors, error; nerrors = 0; while (nchanges > 0) { n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; error = k_ops->k_copyin(k_ops->arg, keva, n); if (error) return (error); changes = keva; for (i = 0; i < n; i++) { kevp = &changes[i]; if (!kevp->filter) continue; kevp->flags &= ~EV_SYSFLAGS; error = kqueue_register(kq, kevp, td, M_WAITOK); if (error || (kevp->flags & EV_RECEIPT)) { if (nevents == 0) return (error); kevp->flags = EV_ERROR; kevp->data = error; (void)k_ops->k_copyout(k_ops->arg, kevp, 1); nevents--; nerrors++; } } nchanges -= n; } if (nerrors) { td->td_retval[0] = nerrors; return (0); } return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td)); } int kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents, struct kevent_copyops *k_ops, const struct timespec *timeout) { struct kqueue *kq; int error; error = kqueue_acquire(fp, &kq); if (error != 0) return (error); error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout); kqueue_release(kq, 0); return (error); } /* * Performs a kevent() call on a temporarily created kqueue. This can be * used to perform one-shot polling, similar to poll() and select(). */ int kern_kevent_anonymous(struct thread *td, int nevents, struct kevent_copyops *k_ops) { struct kqueue kq = {}; int error; kqueue_init(&kq); kq.kq_refcnt = 1; error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL); kqueue_drain(&kq, td); kqueue_destroy(&kq); return (error); } int kqueue_add_filteropts(int filt, struct filterops *filtops) { int error; error = 0; if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { printf( "trying to add a filterop that is out of range: %d is beyond %d\n", ~filt, EVFILT_SYSCOUNT); return EINVAL; } mtx_lock(&filterops_lock); if (sysfilt_ops[~filt].for_fop != &null_filtops && sysfilt_ops[~filt].for_fop != NULL) error = EEXIST; else { sysfilt_ops[~filt].for_fop = filtops; sysfilt_ops[~filt].for_refcnt = 0; } mtx_unlock(&filterops_lock); return (error); } int kqueue_del_filteropts(int filt) { int error; error = 0; if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return EINVAL; mtx_lock(&filterops_lock); if (sysfilt_ops[~filt].for_fop == &null_filtops || sysfilt_ops[~filt].for_fop == NULL) error = EINVAL; else if (sysfilt_ops[~filt].for_refcnt != 0) error = EBUSY; else { sysfilt_ops[~filt].for_fop = &null_filtops; sysfilt_ops[~filt].for_refcnt = 0; } mtx_unlock(&filterops_lock); return error; } static struct filterops * kqueue_fo_find(int filt) { if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return NULL; if (sysfilt_ops[~filt].for_nolock) return sysfilt_ops[~filt].for_fop; mtx_lock(&filterops_lock); sysfilt_ops[~filt].for_refcnt++; if (sysfilt_ops[~filt].for_fop == NULL) sysfilt_ops[~filt].for_fop = &null_filtops; mtx_unlock(&filterops_lock); return sysfilt_ops[~filt].for_fop; } static void kqueue_fo_release(int filt) { if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return; if (sysfilt_ops[~filt].for_nolock) return; mtx_lock(&filterops_lock); KASSERT(sysfilt_ops[~filt].for_refcnt > 0, ("filter object refcount not valid on release")); sysfilt_ops[~filt].for_refcnt--; mtx_unlock(&filterops_lock); } /* * A ref to kq (obtained via kqueue_acquire) must be held. */ static int kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int mflag) { struct filterops *fops; struct file *fp; struct knote *kn, *tkn; struct knlist *knl; int error, filt, event; int haskqglobal, filedesc_unlock; if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE)) return (EINVAL); fp = NULL; kn = NULL; knl = NULL; error = 0; haskqglobal = 0; filedesc_unlock = 0; filt = kev->filter; fops = kqueue_fo_find(filt); if (fops == NULL) return EINVAL; if (kev->flags & EV_ADD) { /* * Prevent waiting with locks. Non-sleepable * allocation failures are handled in the loop, only * if the spare knote appears to be actually required. */ tkn = knote_alloc(mflag); } else { tkn = NULL; } findkn: if (fops->f_isfd) { KASSERT(td != NULL, ("td is NULL")); if (kev->ident > INT_MAX) error = EBADF; else error = fget(td, kev->ident, &cap_event_rights, &fp); if (error) goto done; if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, kev->ident, M_NOWAIT) != 0) { /* try again */ fdrop(fp, td); fp = NULL; error = kqueue_expand(kq, fops, kev->ident, mflag); if (error) goto done; goto findkn; } if (fp->f_type == DTYPE_KQUEUE) { /* * If we add some intelligence about what we are doing, * we should be able to support events on ourselves. * We need to know when we are doing this to prevent * getting both the knlist lock and the kq lock since * they are the same thing. */ if (fp->f_data == kq) { error = EINVAL; goto done; } /* * Pre-lock the filedesc before the global * lock mutex, see the comment in * kqueue_close(). */ FILEDESC_XLOCK(td->td_proc->p_fd); filedesc_unlock = 1; KQ_GLOBAL_LOCK(&kq_global, haskqglobal); } KQ_LOCK(kq); if (kev->ident < kq->kq_knlistsize) { SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) if (kev->filter == kn->kn_filter) break; } } else { if ((kev->flags & EV_ADD) == EV_ADD) { error = kqueue_expand(kq, fops, kev->ident, mflag); if (error != 0) goto done; } KQ_LOCK(kq); /* * If possible, find an existing knote to use for this kevent. */ if (kev->filter == EVFILT_PROC && (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) { /* This is an internal creation of a process tracking * note. Don't attempt to coalesce this with an * existing note. */ ; } else if (kq->kq_knhashmask != 0) { struct klist *list; list = &kq->kq_knhash[ KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; SLIST_FOREACH(kn, list, kn_link) if (kev->ident == kn->kn_id && kev->filter == kn->kn_filter) break; } } /* knote is in the process of changing, wait for it to stabilize. */ if (kn != NULL && kn_in_flux(kn)) { KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (filedesc_unlock) { FILEDESC_XUNLOCK(td->td_proc->p_fd); filedesc_unlock = 0; } kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); if (fp != NULL) { fdrop(fp, td); fp = NULL; } goto findkn; } /* * kn now contains the matching knote, or NULL if no match */ if (kn == NULL) { if (kev->flags & EV_ADD) { kn = tkn; tkn = NULL; if (kn == NULL) { KQ_UNLOCK(kq); error = ENOMEM; goto done; } kn->kn_fp = fp; kn->kn_kq = kq; kn->kn_fop = fops; /* * apply reference counts to knote structure, and * do not release it at the end of this routine. */ fops = NULL; fp = NULL; kn->kn_sfflags = kev->fflags; kn->kn_sdata = kev->data; kev->fflags = 0; kev->data = 0; kn->kn_kevent = *kev; kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT); kn->kn_status = KN_DETACHED; if ((kev->flags & EV_DISABLE) != 0) kn->kn_status |= KN_DISABLED; kn_enter_flux(kn); error = knote_attach(kn, kq); KQ_UNLOCK(kq); if (error != 0) { tkn = kn; goto done; } if ((error = kn->kn_fop->f_attach(kn)) != 0) { knote_drop_detached(kn, td); goto done; } knl = kn_list_lock(kn); goto done_ev_add; } else { /* No matching knote and the EV_ADD flag is not set. */ KQ_UNLOCK(kq); error = ENOENT; goto done; } } if (kev->flags & EV_DELETE) { kn_enter_flux(kn); KQ_UNLOCK(kq); knote_drop(kn, td); goto done; } if (kev->flags & EV_FORCEONESHOT) { kn->kn_flags |= EV_ONESHOT; KNOTE_ACTIVATE(kn, 1); } if ((kev->flags & EV_ENABLE) != 0) kn->kn_status &= ~KN_DISABLED; else if ((kev->flags & EV_DISABLE) != 0) kn->kn_status |= KN_DISABLED; /* * The user may change some filter values after the initial EV_ADD, * but doing so will not reset any filter which has already been * triggered. */ kn->kn_status |= KN_SCAN; kn_enter_flux(kn); KQ_UNLOCK(kq); knl = kn_list_lock(kn); kn->kn_kevent.udata = kev->udata; if (!fops->f_isfd && fops->f_touch != NULL) { fops->f_touch(kn, kev, EVENT_REGISTER); } else { kn->kn_sfflags = kev->fflags; kn->kn_sdata = kev->data; } done_ev_add: /* * We can get here with kn->kn_knlist == NULL. This can happen when * the initial attach event decides that the event is "completed" * already, e.g., filt_procattach() is called on a zombie process. It * will call filt_proc() which will remove it from the list, and NULL * kn_knlist. * * KN_DISABLED will be stable while the knote is in flux, so the * unlocked read will not race with an update. */ if ((kn->kn_status & KN_DISABLED) == 0) event = kn->kn_fop->f_event(kn, 0); else event = 0; KQ_LOCK(kq); if (event) kn->kn_status |= KN_ACTIVE; if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) == KN_ACTIVE) knote_enqueue(kn); kn->kn_status &= ~KN_SCAN; kn_leave_flux(kn); kn_list_unlock(knl); KQ_UNLOCK_FLUX(kq); done: KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (filedesc_unlock) FILEDESC_XUNLOCK(td->td_proc->p_fd); if (fp != NULL) fdrop(fp, td); knote_free(tkn); if (fops != NULL) kqueue_fo_release(filt); return (error); } static int kqueue_acquire(struct file *fp, struct kqueue **kqp) { int error; struct kqueue *kq; error = 0; kq = fp->f_data; if (fp->f_type != DTYPE_KQUEUE || kq == NULL) return (EBADF); *kqp = kq; KQ_LOCK(kq); if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { KQ_UNLOCK(kq); return (EBADF); } kq->kq_refcnt++; KQ_UNLOCK(kq); return error; } static void kqueue_release(struct kqueue *kq, int locked) { if (locked) KQ_OWNED(kq); else KQ_LOCK(kq); kq->kq_refcnt--; if (kq->kq_refcnt == 1) wakeup(&kq->kq_refcnt); if (!locked) KQ_UNLOCK(kq); } static void kqueue_schedtask(struct kqueue *kq) { KQ_OWNED(kq); KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), ("scheduling kqueue task while draining")); if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task); kq->kq_state |= KQ_TASKSCHED; } } /* * Expand the kq to make sure we have storage for fops/ident pair. * * Return 0 on success (or no work necessary), return errno on failure. */ static int kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, int mflag) { struct klist *list, *tmp_knhash, *to_free; u_long tmp_knhashmask; int error, fd, size; KQ_NOTOWNED(kq); error = 0; to_free = NULL; if (fops->f_isfd) { fd = ident; if (kq->kq_knlistsize <= fd) { size = kq->kq_knlistsize; while (size <= fd) size += KQEXTENT; list = malloc(size * sizeof(*list), M_KQUEUE, mflag); if (list == NULL) return ENOMEM; KQ_LOCK(kq); if ((kq->kq_state & KQ_CLOSING) != 0) { to_free = list; error = EBADF; } else if (kq->kq_knlistsize > fd) { to_free = list; } else { if (kq->kq_knlist != NULL) { bcopy(kq->kq_knlist, list, kq->kq_knlistsize * sizeof(*list)); to_free = kq->kq_knlist; kq->kq_knlist = NULL; } bzero((caddr_t)list + kq->kq_knlistsize * sizeof(*list), (size - kq->kq_knlistsize) * sizeof(*list)); kq->kq_knlistsize = size; kq->kq_knlist = list; } KQ_UNLOCK(kq); } } else { if (kq->kq_knhashmask == 0) { tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE, &tmp_knhashmask, (mflag & M_WAITOK) != 0 ? HASH_WAITOK : HASH_NOWAIT); if (tmp_knhash == NULL) return (ENOMEM); KQ_LOCK(kq); if ((kq->kq_state & KQ_CLOSING) != 0) { to_free = tmp_knhash; error = EBADF; } else if (kq->kq_knhashmask == 0) { kq->kq_knhash = tmp_knhash; kq->kq_knhashmask = tmp_knhashmask; } else { to_free = tmp_knhash; } KQ_UNLOCK(kq); } } free(to_free, M_KQUEUE); KQ_NOTOWNED(kq); return (error); } static void kqueue_task(void *arg, int pending) { struct kqueue *kq; int haskqglobal; haskqglobal = 0; kq = arg; KQ_GLOBAL_LOCK(&kq_global, haskqglobal); KQ_LOCK(kq); KNOTE_LOCKED(&kq->kq_sel.si_note, 0); kq->kq_state &= ~KQ_TASKSCHED; if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { wakeup(&kq->kq_state); } KQ_UNLOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); } /* * Scan, update kn_data (if not ONESHOT), and copyout triggered events. * We treat KN_MARKER knotes as if they are in flux. */ static int kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, const struct timespec *tsp, struct kevent *keva, struct thread *td) { struct kevent *kevp; struct knote *kn, *marker; struct knlist *knl; sbintime_t asbt, rsbt; int count, error, haskqglobal, influx, nkev, touch; count = maxevents; nkev = 0; error = 0; haskqglobal = 0; if (maxevents == 0) goto done_nl; rsbt = 0; if (tsp != NULL) { if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000) { error = EINVAL; goto done_nl; } if (timespecisset(tsp)) { if (tsp->tv_sec <= INT32_MAX) { rsbt = tstosbt(*tsp); if (TIMESEL(&asbt, rsbt)) asbt += tc_tick_sbt; if (asbt <= SBT_MAX - rsbt) asbt += rsbt; else asbt = 0; rsbt >>= tc_precexp; } else asbt = 0; } else asbt = -1; } else asbt = 0; marker = knote_alloc(M_WAITOK); marker->kn_status = KN_MARKER; KQ_LOCK(kq); retry: kevp = keva; if (kq->kq_count == 0) { if (asbt == -1) { error = EWOULDBLOCK; } else { kq->kq_state |= KQ_SLEEP; error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, "kqread", asbt, rsbt, C_ABSOLUTE); } if (error == 0) goto retry; /* don't restart after signals... */ if (error == ERESTART) error = EINTR; else if (error == EWOULDBLOCK) error = 0; goto done; } TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); influx = 0; while (count) { KQ_OWNED(kq); kn = TAILQ_FIRST(&kq->kq_head); if ((kn->kn_status == KN_MARKER && kn != marker) || kn_in_flux(kn)) { if (influx) { influx = 0; KQ_FLUX_WAKEUP(kq); } kq->kq_state |= KQ_FLUXWAIT; error = msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); continue; } TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { kn->kn_status &= ~KN_QUEUED; kq->kq_count--; continue; } if (kn == marker) { KQ_FLUX_WAKEUP(kq); if (count == maxevents) goto retry; goto done; } KASSERT(!kn_in_flux(kn), ("knote %p is unexpectedly in flux", kn)); if ((kn->kn_flags & EV_DROP) == EV_DROP) { kn->kn_status &= ~KN_QUEUED; kn_enter_flux(kn); kq->kq_count--; KQ_UNLOCK(kq); /* * We don't need to lock the list since we've * marked it as in flux. */ knote_drop(kn, td); KQ_LOCK(kq); continue; } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { kn->kn_status &= ~KN_QUEUED; kn_enter_flux(kn); kq->kq_count--; KQ_UNLOCK(kq); /* * We don't need to lock the list since we've * marked the knote as being in flux. */ *kevp = kn->kn_kevent; knote_drop(kn, td); KQ_LOCK(kq); kn = NULL; } else { kn->kn_status |= KN_SCAN; kn_enter_flux(kn); KQ_UNLOCK(kq); if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) KQ_GLOBAL_LOCK(&kq_global, haskqglobal); knl = kn_list_lock(kn); if (kn->kn_fop->f_event(kn, 0) == 0) { KQ_LOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE | KN_SCAN); kn_leave_flux(kn); kq->kq_count--; kn_list_unlock(knl); influx = 1; continue; } touch = (!kn->kn_fop->f_isfd && kn->kn_fop->f_touch != NULL); if (touch) kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); else *kevp = kn->kn_kevent; KQ_LOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { /* * Manually clear knotes who weren't * 'touch'ed. */ if (touch == 0 && kn->kn_flags & EV_CLEAR) { kn->kn_data = 0; kn->kn_fflags = 0; } if (kn->kn_flags & EV_DISPATCH) kn->kn_status |= KN_DISABLED; kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); kq->kq_count--; } else TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); kn->kn_status &= ~KN_SCAN; kn_leave_flux(kn); kn_list_unlock(knl); influx = 1; } /* we are returning a copy to the user */ kevp++; nkev++; count--; if (nkev == KQ_NEVENTS) { influx = 0; KQ_UNLOCK_FLUX(kq); error = k_ops->k_copyout(k_ops->arg, keva, nkev); nkev = 0; kevp = keva; KQ_LOCK(kq); if (error) break; } } TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); done: KQ_OWNED(kq); KQ_UNLOCK_FLUX(kq); knote_free(marker); done_nl: KQ_NOTOWNED(kq); if (nkev != 0) error = k_ops->k_copyout(k_ops->arg, keva, nkev); td->td_retval[0] = maxevents - count; return (error); } /*ARGSUSED*/ static int kqueue_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, struct thread *td) { /* * Enabling sigio causes two major problems: * 1) infinite recursion: * Synopsys: kevent is being used to track signals and have FIOASYNC * set. On receipt of a signal this will cause a kqueue to recurse * into itself over and over. Sending the sigio causes the kqueue * to become ready, which in turn posts sigio again, forever. * Solution: this can be solved by setting a flag in the kqueue that * we have a SIGIO in progress. * 2) locking problems: * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts * us above the proc and pgrp locks. * Solution: Post a signal using an async mechanism, being sure to * record a generation count in the delivery so that we do not deliver * a signal to the wrong process. * * Note, these two mechanisms are somewhat mutually exclusive! */ #if 0 struct kqueue *kq; kq = fp->f_data; switch (cmd) { case FIOASYNC: if (*(int *)data) { kq->kq_state |= KQ_ASYNC; } else { kq->kq_state &= ~KQ_ASYNC; } return (0); case FIOSETOWN: return (fsetown(*(int *)data, &kq->kq_sigio)); case FIOGETOWN: *(int *)data = fgetown(&kq->kq_sigio); return (0); } #endif return (ENOTTY); } /*ARGSUSED*/ static int kqueue_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct kqueue *kq; int revents = 0; int error; if ((error = kqueue_acquire(fp, &kq))) return POLLERR; KQ_LOCK(kq); if (events & (POLLIN | POLLRDNORM)) { if (kq->kq_count) { revents |= events & (POLLIN | POLLRDNORM); } else { selrecord(td, &kq->kq_sel); if (SEL_WAITING(&kq->kq_sel)) kq->kq_state |= KQ_SEL; } } kqueue_release(kq, 1); KQ_UNLOCK(kq); return (revents); } /*ARGSUSED*/ static int kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, struct thread *td) { bzero((void *)st, sizeof *st); /* * We no longer return kq_count because the unlocked value is useless. * If you spent all this time getting the count, why not spend your * syscall better by calling kevent? * * XXX - This is needed for libc_r. */ st->st_mode = S_IFIFO; return (0); } static void kqueue_drain(struct kqueue *kq, struct thread *td) { struct knote *kn; int i; KQ_LOCK(kq); KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, ("kqueue already closing")); kq->kq_state |= KQ_CLOSING; if (kq->kq_refcnt > 1) msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); KASSERT(knlist_empty(&kq->kq_sel.si_note), ("kqueue's knlist not empty")); for (i = 0; i < kq->kq_knlistsize; i++) { while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { if (kn_in_flux(kn)) { kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); continue; } kn_enter_flux(kn); KQ_UNLOCK(kq); knote_drop(kn, td); KQ_LOCK(kq); } } if (kq->kq_knhashmask != 0) { for (i = 0; i <= kq->kq_knhashmask; i++) { while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { if (kn_in_flux(kn)) { kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqclo2", 0); continue; } kn_enter_flux(kn); KQ_UNLOCK(kq); knote_drop(kn, td); KQ_LOCK(kq); } } } if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { kq->kq_state |= KQ_TASKDRAIN; msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); } if ((kq->kq_state & KQ_SEL) == KQ_SEL) { selwakeuppri(&kq->kq_sel, PSOCK); if (!SEL_WAITING(&kq->kq_sel)) kq->kq_state &= ~KQ_SEL; } KQ_UNLOCK(kq); } static void kqueue_destroy(struct kqueue *kq) { KASSERT(kq->kq_fdp == NULL, ("kqueue still attached to a file descriptor")); seldrain(&kq->kq_sel); knlist_destroy(&kq->kq_sel.si_note); mtx_destroy(&kq->kq_lock); if (kq->kq_knhash != NULL) free(kq->kq_knhash, M_KQUEUE); if (kq->kq_knlist != NULL) free(kq->kq_knlist, M_KQUEUE); funsetown(&kq->kq_sigio); } /*ARGSUSED*/ static int kqueue_close(struct file *fp, struct thread *td) { struct kqueue *kq = fp->f_data; struct filedesc *fdp; int error; int filedesc_unlock; if ((error = kqueue_acquire(fp, &kq))) return error; kqueue_drain(kq, td); /* * We could be called due to the knote_drop() doing fdrop(), * called from kqueue_register(). In this case the global * lock is owned, and filedesc sx is locked before, to not * take the sleepable lock after non-sleepable. */ fdp = kq->kq_fdp; kq->kq_fdp = NULL; if (!sx_xlocked(FILEDESC_LOCK(fdp))) { FILEDESC_XLOCK(fdp); filedesc_unlock = 1; } else filedesc_unlock = 0; TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); if (filedesc_unlock) FILEDESC_XUNLOCK(fdp); kqueue_destroy(kq); chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0); crfree(kq->kq_cred); free(kq, M_KQUEUE); fp->f_data = NULL; return (0); } static int kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { kif->kf_type = KF_TYPE_KQUEUE; return (0); } static void kqueue_wakeup(struct kqueue *kq) { KQ_OWNED(kq); if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { kq->kq_state &= ~KQ_SLEEP; wakeup(kq); } if ((kq->kq_state & KQ_SEL) == KQ_SEL) { selwakeuppri(&kq->kq_sel, PSOCK); if (!SEL_WAITING(&kq->kq_sel)) kq->kq_state &= ~KQ_SEL; } if (!knlist_empty(&kq->kq_sel.si_note)) kqueue_schedtask(kq); if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { pgsigio(&kq->kq_sigio, SIGIO, 0); } } /* * Walk down a list of knotes, activating them if their event has triggered. * * There is a possibility to optimize in the case of one kq watching another. * Instead of scheduling a task to wake it up, you could pass enough state * down the chain to make up the parent kqueue. Make this code functional * first. */ void knote(struct knlist *list, long hint, int lockflags) { struct kqueue *kq; struct knote *kn, *tkn; int error; if (list == NULL) return; KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); if ((lockflags & KNF_LISTLOCKED) == 0) list->kl_lock(list->kl_lockarg); /* * If we unlock the list lock (and enter influx), we can * eliminate the kqueue scheduling, but this will introduce * four lock/unlock's for each knote to test. Also, marker * would be needed to keep iteration position, since filters * or other threads could remove events. */ SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) { kq = kn->kn_kq; KQ_LOCK(kq); if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { /* * Do not process the influx notes, except for * the influx coming from the kq unlock in the * kqueue_scan(). In the later case, we do * not interfere with the scan, since the code * fragment in kqueue_scan() locks the knlist, * and cannot proceed until we finished. */ KQ_UNLOCK(kq); } else if ((lockflags & KNF_NOKQLOCK) != 0) { kn_enter_flux(kn); KQ_UNLOCK(kq); error = kn->kn_fop->f_event(kn, hint); KQ_LOCK(kq); kn_leave_flux(kn); if (error) KNOTE_ACTIVATE(kn, 1); KQ_UNLOCK_FLUX(kq); } else { if (kn->kn_fop->f_event(kn, hint)) KNOTE_ACTIVATE(kn, 1); KQ_UNLOCK(kq); } } if ((lockflags & KNF_LISTLOCKED) == 0) list->kl_unlock(list->kl_lockarg); } /* * add a knote to a knlist */ void knlist_add(struct knlist *knl, struct knote *kn, int islocked) { KNL_ASSERT_LOCK(knl, islocked); KQ_NOTOWNED(kn->kn_kq); KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn)); KASSERT((kn->kn_status & KN_DETACHED) != 0, ("knote %p was not detached", kn)); if (!islocked) knl->kl_lock(knl->kl_lockarg); SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); if (!islocked) knl->kl_unlock(knl->kl_lockarg); KQ_LOCK(kn->kn_kq); kn->kn_knlist = knl; kn->kn_status &= ~KN_DETACHED; KQ_UNLOCK(kn->kn_kq); } static void knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked) { KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked")); KNL_ASSERT_LOCK(knl, knlislocked); mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn)); KASSERT((kn->kn_status & KN_DETACHED) == 0, ("knote %p was already detached", kn)); if (!knlislocked) knl->kl_lock(knl->kl_lockarg); SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); kn->kn_knlist = NULL; if (!knlislocked) kn_list_unlock(knl); if (!kqislocked) KQ_LOCK(kn->kn_kq); kn->kn_status |= KN_DETACHED; if (!kqislocked) KQ_UNLOCK(kn->kn_kq); } /* * remove knote from the specified knlist */ void knlist_remove(struct knlist *knl, struct knote *kn, int islocked) { knlist_remove_kq(knl, kn, islocked, 0); } int knlist_empty(struct knlist *knl) { KNL_ASSERT_LOCKED(knl); return (SLIST_EMPTY(&knl->kl_list)); } static struct mtx knlist_lock; MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", MTX_DEF); static void knlist_mtx_lock(void *arg); static void knlist_mtx_unlock(void *arg); static void knlist_mtx_lock(void *arg) { mtx_lock((struct mtx *)arg); } static void knlist_mtx_unlock(void *arg) { mtx_unlock((struct mtx *)arg); } static void -knlist_mtx_assert_locked(void *arg) +knlist_mtx_assert_lock(void *arg, int what) { - mtx_assert((struct mtx *)arg, MA_OWNED); -} - -static void -knlist_mtx_assert_unlocked(void *arg) -{ - - mtx_assert((struct mtx *)arg, MA_NOTOWNED); + if (what == LA_LOCKED) + mtx_assert((struct mtx *)arg, MA_OWNED); + else + mtx_assert((struct mtx *)arg, MA_NOTOWNED); } static void knlist_rw_rlock(void *arg) { rw_rlock((struct rwlock *)arg); } static void knlist_rw_runlock(void *arg) { rw_runlock((struct rwlock *)arg); } static void -knlist_rw_assert_locked(void *arg) +knlist_rw_assert_lock(void *arg, int what) { - rw_assert((struct rwlock *)arg, RA_LOCKED); -} - -static void -knlist_rw_assert_unlocked(void *arg) -{ - - rw_assert((struct rwlock *)arg, RA_UNLOCKED); + if (what == LA_LOCKED) + rw_assert((struct rwlock *)arg, RA_LOCKED); + else + rw_assert((struct rwlock *)arg, RA_UNLOCKED); } void knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), void (*kl_unlock)(void *), - void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *)) + void (*kl_assert_lock)(void *, int)) { if (lock == NULL) knl->kl_lockarg = &knlist_lock; else knl->kl_lockarg = lock; if (kl_lock == NULL) knl->kl_lock = knlist_mtx_lock; else knl->kl_lock = kl_lock; if (kl_unlock == NULL) knl->kl_unlock = knlist_mtx_unlock; else knl->kl_unlock = kl_unlock; - if (kl_assert_locked == NULL) - knl->kl_assert_locked = knlist_mtx_assert_locked; - else - knl->kl_assert_locked = kl_assert_locked; - if (kl_assert_unlocked == NULL) - knl->kl_assert_unlocked = knlist_mtx_assert_unlocked; + if (kl_assert_lock == NULL) + knl->kl_assert_lock = knlist_mtx_assert_lock; else - knl->kl_assert_unlocked = kl_assert_unlocked; + knl->kl_assert_lock = kl_assert_lock; knl->kl_autodestroy = 0; SLIST_INIT(&knl->kl_list); } void knlist_init_mtx(struct knlist *knl, struct mtx *lock) { - knlist_init(knl, lock, NULL, NULL, NULL, NULL); + knlist_init(knl, lock, NULL, NULL, NULL); } struct knlist * knlist_alloc(struct mtx *lock) { struct knlist *knl; knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK); knlist_init_mtx(knl, lock); return (knl); } void knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock) { knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock, - knlist_rw_assert_locked, knlist_rw_assert_unlocked); + knlist_rw_assert_lock); } void knlist_destroy(struct knlist *knl) { KASSERT(KNLIST_EMPTY(knl), ("destroying knlist %p with knotes on it", knl)); } void knlist_detach(struct knlist *knl) { KNL_ASSERT_LOCKED(knl); knl->kl_autodestroy = 1; if (knlist_empty(knl)) { knlist_destroy(knl); free(knl, M_KQUEUE); } } /* * Even if we are locked, we may need to drop the lock to allow any influx * knotes time to "settle". */ void knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) { struct knote *kn, *kn2; struct kqueue *kq; KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl)); if (islocked) KNL_ASSERT_LOCKED(knl); else { KNL_ASSERT_UNLOCKED(knl); again: /* need to reacquire lock since we have dropped it */ knl->kl_lock(knl->kl_lockarg); } SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { kq = kn->kn_kq; KQ_LOCK(kq); if (kn_in_flux(kn)) { KQ_UNLOCK(kq); continue; } knlist_remove_kq(knl, kn, 1, 1); if (killkn) { kn_enter_flux(kn); KQ_UNLOCK(kq); knote_drop_detached(kn, td); } else { /* Make sure cleared knotes disappear soon */ kn->kn_flags |= EV_EOF | EV_ONESHOT; KQ_UNLOCK(kq); } kq = NULL; } if (!SLIST_EMPTY(&knl->kl_list)) { /* there are still in flux knotes remaining */ kn = SLIST_FIRST(&knl->kl_list); kq = kn->kn_kq; KQ_LOCK(kq); KASSERT(kn_in_flux(kn), ("knote removed w/o list lock")); knl->kl_unlock(knl->kl_lockarg); kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); kq = NULL; goto again; } if (islocked) KNL_ASSERT_LOCKED(knl); else { knl->kl_unlock(knl->kl_lockarg); KNL_ASSERT_UNLOCKED(knl); } } /* * Remove all knotes referencing a specified fd must be called with FILEDESC * lock. This prevents a race where a new fd comes along and occupies the * entry and we attach a knote to the fd. */ void knote_fdclose(struct thread *td, int fd) { struct filedesc *fdp = td->td_proc->p_fd; struct kqueue *kq; struct knote *kn; int influx; FILEDESC_XLOCK_ASSERT(fdp); /* * We shouldn't have to worry about new kevents appearing on fd * since filedesc is locked. */ TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { KQ_LOCK(kq); again: influx = 0; while (kq->kq_knlistsize > fd && (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { if (kn_in_flux(kn)) { /* someone else might be waiting on our knote */ if (influx) wakeup(kq); kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); goto again; } kn_enter_flux(kn); KQ_UNLOCK(kq); influx = 1; knote_drop(kn, td); KQ_LOCK(kq); } KQ_UNLOCK_FLUX(kq); } } static int knote_attach(struct knote *kn, struct kqueue *kq) { struct klist *list; KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn)); KQ_OWNED(kq); if ((kq->kq_state & KQ_CLOSING) != 0) return (EBADF); if (kn->kn_fop->f_isfd) { if (kn->kn_id >= kq->kq_knlistsize) return (ENOMEM); list = &kq->kq_knlist[kn->kn_id]; } else { if (kq->kq_knhash == NULL) return (ENOMEM); list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; } SLIST_INSERT_HEAD(list, kn, kn_link); return (0); } static void knote_drop(struct knote *kn, struct thread *td) { if ((kn->kn_status & KN_DETACHED) == 0) kn->kn_fop->f_detach(kn); knote_drop_detached(kn, td); } static void knote_drop_detached(struct knote *kn, struct thread *td) { struct kqueue *kq; struct klist *list; kq = kn->kn_kq; KASSERT((kn->kn_status & KN_DETACHED) != 0, ("knote %p still attached", kn)); KQ_NOTOWNED(kq); KQ_LOCK(kq); KASSERT(kn->kn_influx == 1, ("knote_drop called on %p with influx %d", kn, kn->kn_influx)); if (kn->kn_fop->f_isfd) list = &kq->kq_knlist[kn->kn_id]; else list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; if (!SLIST_EMPTY(list)) SLIST_REMOVE(list, kn, knote, kn_link); if (kn->kn_status & KN_QUEUED) knote_dequeue(kn); KQ_UNLOCK_FLUX(kq); if (kn->kn_fop->f_isfd) { fdrop(kn->kn_fp, td); kn->kn_fp = NULL; } kqueue_fo_release(kn->kn_kevent.filter); kn->kn_fop = NULL; knote_free(kn); } static void knote_enqueue(struct knote *kn) { struct kqueue *kq = kn->kn_kq; KQ_OWNED(kn->kn_kq); KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); kn->kn_status |= KN_QUEUED; kq->kq_count++; kqueue_wakeup(kq); } static void knote_dequeue(struct knote *kn) { struct kqueue *kq = kn->kn_kq; KQ_OWNED(kn->kn_kq); KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); kn->kn_status &= ~KN_QUEUED; kq->kq_count--; } static void knote_init(void) { knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); } SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); static struct knote * knote_alloc(int mflag) { return (uma_zalloc(knote_zone, mflag | M_ZERO)); } static void knote_free(struct knote *kn) { uma_zfree(knote_zone, kn); } /* * Register the kev w/ the kq specified by fd. */ int kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag) { struct kqueue *kq; struct file *fp; cap_rights_t rights; int error; error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp); if (error != 0) return (error); if ((error = kqueue_acquire(fp, &kq)) != 0) goto noacquire; error = kqueue_register(kq, kev, td, mflag); kqueue_release(kq, 0); noacquire: fdrop(fp, td); return (error); } diff --git a/sys/kern/uipc_socket.c b/sys/kern/uipc_socket.c index 77ca888d30f0..32fc8f3d4c7d 100644 --- a/sys/kern/uipc_socket.c +++ b/sys/kern/uipc_socket.c @@ -1,4424 +1,4414 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993 * The Regents of the University of California. * Copyright (c) 2004 The FreeBSD Foundation * Copyright (c) 2004-2008 Robert N. M. Watson * 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. * 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. * * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 */ /* * Comments on the socket life cycle: * * soalloc() sets of socket layer state for a socket, called only by * socreate() and sonewconn(). Socket layer private. * * sodealloc() tears down socket layer state for a socket, called only by * sofree() and sonewconn(). Socket layer private. * * pru_attach() associates protocol layer state with an allocated socket; * called only once, may fail, aborting socket allocation. This is called * from socreate() and sonewconn(). Socket layer private. * * pru_detach() disassociates protocol layer state from an attached socket, * and will be called exactly once for sockets in which pru_attach() has * been successfully called. If pru_attach() returned an error, * pru_detach() will not be called. Socket layer private. * * pru_abort() and pru_close() notify the protocol layer that the last * consumer of a socket is starting to tear down the socket, and that the * protocol should terminate the connection. Historically, pru_abort() also * detached protocol state from the socket state, but this is no longer the * case. * * socreate() creates a socket and attaches protocol state. This is a public * interface that may be used by socket layer consumers to create new * sockets. * * sonewconn() creates a socket and attaches protocol state. This is a * public interface that may be used by protocols to create new sockets when * a new connection is received and will be available for accept() on a * listen socket. * * soclose() destroys a socket after possibly waiting for it to disconnect. * This is a public interface that socket consumers should use to close and * release a socket when done with it. * * soabort() destroys a socket without waiting for it to disconnect (used * only for incoming connections that are already partially or fully * connected). This is used internally by the socket layer when clearing * listen socket queues (due to overflow or close on the listen socket), but * is also a public interface protocols may use to abort connections in * their incomplete listen queues should they no longer be required. Sockets * placed in completed connection listen queues should not be aborted for * reasons described in the comment above the soclose() implementation. This * is not a general purpose close routine, and except in the specific * circumstances described here, should not be used. * * sofree() will free a socket and its protocol state if all references on * the socket have been released, and is the public interface to attempt to * free a socket when a reference is removed. This is a socket layer private * interface. * * NOTE: In addition to socreate() and soclose(), which provide a single * socket reference to the consumer to be managed as required, there are two * calls to explicitly manage socket references, soref(), and sorele(). * Currently, these are generally required only when transitioning a socket * from a listen queue to a file descriptor, in order to prevent garbage * collection of the socket at an untimely moment. For a number of reasons, * these interfaces are not preferred, and should be avoided. * * NOTE: With regard to VNETs the general rule is that callers do not set * curvnet. Exceptions to this rule include soabort(), sodisconnect(), * sofree() (and with that sorele(), sotryfree()), as well as sonewconn() * and sorflush(), which are usually called from a pre-set VNET context. * sopoll() currently does not need a VNET context to be set. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_kern_tls.h" #include "opt_sctp.h" #include #include #include #include #include #include #include #include #include #include #include /* for struct knote */ #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 COMPAT_FREEBSD32 #include #include #include #endif static int soreceive_rcvoob(struct socket *so, struct uio *uio, int flags); static void so_rdknl_lock(void *); static void so_rdknl_unlock(void *); -static void so_rdknl_assert_locked(void *); -static void so_rdknl_assert_unlocked(void *); +static void so_rdknl_assert_lock(void *, int); static void so_wrknl_lock(void *); static void so_wrknl_unlock(void *); -static void so_wrknl_assert_locked(void *); -static void so_wrknl_assert_unlocked(void *); +static void so_wrknl_assert_lock(void *, int); static void filt_sordetach(struct knote *kn); static int filt_soread(struct knote *kn, long hint); static void filt_sowdetach(struct knote *kn); static int filt_sowrite(struct knote *kn, long hint); static int filt_soempty(struct knote *kn, long hint); static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id); fo_kqfilter_t soo_kqfilter; static struct filterops soread_filtops = { .f_isfd = 1, .f_detach = filt_sordetach, .f_event = filt_soread, }; static struct filterops sowrite_filtops = { .f_isfd = 1, .f_detach = filt_sowdetach, .f_event = filt_sowrite, }; static struct filterops soempty_filtops = { .f_isfd = 1, .f_detach = filt_sowdetach, .f_event = filt_soempty, }; so_gen_t so_gencnt; /* generation count for sockets */ MALLOC_DEFINE(M_SONAME, "soname", "socket name"); MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); #define VNET_SO_ASSERT(so) \ VNET_ASSERT(curvnet != NULL, \ ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); #define V_socket_hhh VNET(socket_hhh) /* * Limit on the number of connections in the listen queue waiting * for accept(2). * NB: The original sysctl somaxconn is still available but hidden * to prevent confusion about the actual purpose of this number. */ static u_int somaxconn = SOMAXCONN; static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS) { int error; int val; val = somaxconn; error = sysctl_handle_int(oidp, &val, 0, req); if (error || !req->newptr ) return (error); /* * The purpose of the UINT_MAX / 3 limit, is so that the formula * 3 * so_qlimit / 2 * below, will not overflow. */ if (val < 1 || val > UINT_MAX / 3) return (EINVAL); somaxconn = val; return (0); } SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int), sysctl_somaxconn, "I", "Maximum listen socket pending connection accept queue size"); SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0, sizeof(int), sysctl_somaxconn, "I", "Maximum listen socket pending connection accept queue size (compat)"); static int numopensockets; SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, &numopensockets, 0, "Number of open sockets"); /* * accept_mtx locks down per-socket fields relating to accept queues. See * socketvar.h for an annotation of the protected fields of struct socket. */ struct mtx accept_mtx; MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); /* * so_global_mtx protects so_gencnt, numopensockets, and the per-socket * so_gencnt field. */ static struct mtx so_global_mtx; MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); /* * General IPC sysctl name space, used by sockets and a variety of other IPC * types. */ SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "IPC"); /* * Initialize the socket subsystem and set up the socket * memory allocator. */ static uma_zone_t socket_zone; int maxsockets; static void socket_zone_change(void *tag) { maxsockets = uma_zone_set_max(socket_zone, maxsockets); } static void socket_hhook_register(int subtype) { if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, &V_socket_hhh[subtype], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register hook\n", __func__); } static void socket_hhook_deregister(int subtype) { if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) printf("%s: WARNING: unable to deregister hook\n", __func__); } static void socket_init(void *tag) { socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); maxsockets = uma_zone_set_max(socket_zone, maxsockets); uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, EVENTHANDLER_PRI_FIRST); } SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); static void socket_vnet_init(const void *unused __unused) { int i; /* We expect a contiguous range */ for (i = 0; i <= HHOOK_SOCKET_LAST; i++) socket_hhook_register(i); } VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_vnet_init, NULL); static void socket_vnet_uninit(const void *unused __unused) { int i; for (i = 0; i <= HHOOK_SOCKET_LAST; i++) socket_hhook_deregister(i); } VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_vnet_uninit, NULL); /* * Initialise maxsockets. This SYSINIT must be run after * tunable_mbinit(). */ static void init_maxsockets(void *ignored) { TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); maxsockets = imax(maxsockets, maxfiles); } SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); /* * Sysctl to get and set the maximum global sockets limit. Notify protocols * of the change so that they can update their dependent limits as required. */ static int sysctl_maxsockets(SYSCTL_HANDLER_ARGS) { int error, newmaxsockets; newmaxsockets = maxsockets; error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); if (error == 0 && req->newptr) { if (newmaxsockets > maxsockets && newmaxsockets <= maxfiles) { maxsockets = newmaxsockets; EVENTHANDLER_INVOKE(maxsockets_change); } else error = EINVAL; } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0, sysctl_maxsockets, "IU", "Maximum number of sockets available"); /* * Socket operation routines. These routines are called by the routines in * sys_socket.c or from a system process, and implement the semantics of * socket operations by switching out to the protocol specific routines. */ /* * Get a socket structure from our zone, and initialize it. Note that it * would probably be better to allocate socket and PCB at the same time, but * I'm not convinced that all the protocols can be easily modified to do * this. * * soalloc() returns a socket with a ref count of 0. */ static struct socket * soalloc(struct vnet *vnet) { struct socket *so; so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); if (so == NULL) return (NULL); #ifdef MAC if (mac_socket_init(so, M_NOWAIT) != 0) { uma_zfree(socket_zone, so); return (NULL); } #endif if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { uma_zfree(socket_zone, so); return (NULL); } /* * The socket locking protocol allows to lock 2 sockets at a time, * however, the first one must be a listening socket. WITNESS lacks * a feature to change class of an existing lock, so we use DUPOK. */ mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK); SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); so->so_rcv.sb_sel = &so->so_rdsel; so->so_snd.sb_sel = &so->so_wrsel; sx_init(&so->so_snd.sb_sx, "so_snd_sx"); sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); TAILQ_INIT(&so->so_snd.sb_aiojobq); TAILQ_INIT(&so->so_rcv.sb_aiojobq); TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); #ifdef VIMAGE VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", __func__, __LINE__, so)); so->so_vnet = vnet; #endif /* We shouldn't need the so_global_mtx */ if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { /* Do we need more comprehensive error returns? */ uma_zfree(socket_zone, so); return (NULL); } mtx_lock(&so_global_mtx); so->so_gencnt = ++so_gencnt; ++numopensockets; #ifdef VIMAGE vnet->vnet_sockcnt++; #endif mtx_unlock(&so_global_mtx); return (so); } /* * Free the storage associated with a socket at the socket layer, tear down * locks, labels, etc. All protocol state is assumed already to have been * torn down (and possibly never set up) by the caller. */ static void sodealloc(struct socket *so) { KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); mtx_lock(&so_global_mtx); so->so_gencnt = ++so_gencnt; --numopensockets; /* Could be below, but faster here. */ #ifdef VIMAGE VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", __func__, __LINE__, so)); so->so_vnet->vnet_sockcnt--; #endif mtx_unlock(&so_global_mtx); #ifdef MAC mac_socket_destroy(so); #endif hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); crfree(so->so_cred); khelp_destroy_osd(&so->osd); if (SOLISTENING(so)) { if (so->sol_accept_filter != NULL) accept_filt_setopt(so, NULL); } else { if (so->so_rcv.sb_hiwat) (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); if (so->so_snd.sb_hiwat) (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); sx_destroy(&so->so_snd.sb_sx); sx_destroy(&so->so_rcv.sb_sx); SOCKBUF_LOCK_DESTROY(&so->so_snd); SOCKBUF_LOCK_DESTROY(&so->so_rcv); } mtx_destroy(&so->so_lock); uma_zfree(socket_zone, so); } /* * socreate returns a socket with a ref count of 1. The socket should be * closed with soclose(). */ int socreate(int dom, struct socket **aso, int type, int proto, struct ucred *cred, struct thread *td) { struct protosw *prp; struct socket *so; int error; if (proto) prp = pffindproto(dom, proto, type); else prp = pffindtype(dom, type); if (prp == NULL) { /* No support for domain. */ if (pffinddomain(dom) == NULL) return (EAFNOSUPPORT); /* No support for socket type. */ if (proto == 0 && type != 0) return (EPROTOTYPE); return (EPROTONOSUPPORT); } if (prp->pr_usrreqs->pru_attach == NULL || prp->pr_usrreqs->pru_attach == pru_attach_notsupp) return (EPROTONOSUPPORT); if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) return (EPROTONOSUPPORT); if (prp->pr_type != type) return (EPROTOTYPE); so = soalloc(CRED_TO_VNET(cred)); if (so == NULL) return (ENOBUFS); so->so_type = type; so->so_cred = crhold(cred); if ((prp->pr_domain->dom_family == PF_INET) || (prp->pr_domain->dom_family == PF_INET6) || (prp->pr_domain->dom_family == PF_ROUTE)) so->so_fibnum = td->td_proc->p_fibnum; else so->so_fibnum = 0; so->so_proto = prp; #ifdef MAC mac_socket_create(cred, so); #endif knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, - so_rdknl_assert_locked, so_rdknl_assert_unlocked); + so_rdknl_assert_lock); knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, - so_wrknl_assert_locked, so_wrknl_assert_unlocked); + so_wrknl_assert_lock); /* * Auto-sizing of socket buffers is managed by the protocols and * the appropriate flags must be set in the pru_attach function. */ CURVNET_SET(so->so_vnet); error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); CURVNET_RESTORE(); if (error) { sodealloc(so); return (error); } soref(so); *aso = so; return (0); } #ifdef REGRESSION static int regression_sonewconn_earlytest = 1; SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); #endif static struct timeval overinterval = { 60, 0 }; SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW, &overinterval, "Delay in seconds between warnings for listen socket overflows"); /* * When an attempt at a new connection is noted on a socket which accepts * connections, sonewconn is called. If the connection is possible (subject * to space constraints, etc.) then we allocate a new structure, properly * linked into the data structure of the original socket, and return this. * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. * * Note: the ref count on the socket is 0 on return. */ struct socket * sonewconn(struct socket *head, int connstatus) { struct sbuf descrsb; struct socket *so; int len, overcount; u_int qlen; const char localprefix[] = "local:"; char descrbuf[SUNPATHLEN + sizeof(localprefix)]; #if defined(INET6) char addrbuf[INET6_ADDRSTRLEN]; #elif defined(INET) char addrbuf[INET_ADDRSTRLEN]; #endif bool dolog, over; SOLISTEN_LOCK(head); over = (head->sol_qlen > 3 * head->sol_qlimit / 2); #ifdef REGRESSION if (regression_sonewconn_earlytest && over) { #else if (over) { #endif head->sol_overcount++; dolog = !!ratecheck(&head->sol_lastover, &overinterval); /* * If we're going to log, copy the overflow count and queue * length from the listen socket before dropping the lock. * Also, reset the overflow count. */ if (dolog) { overcount = head->sol_overcount; head->sol_overcount = 0; qlen = head->sol_qlen; } SOLISTEN_UNLOCK(head); if (dolog) { /* * Try to print something descriptive about the * socket for the error message. */ sbuf_new(&descrsb, descrbuf, sizeof(descrbuf), SBUF_FIXEDLEN); switch (head->so_proto->pr_domain->dom_family) { #if defined(INET) || defined(INET6) #ifdef INET case AF_INET: #endif #ifdef INET6 case AF_INET6: if (head->so_proto->pr_domain->dom_family == AF_INET6 || (sotoinpcb(head)->inp_inc.inc_flags & INC_ISIPV6)) { ip6_sprintf(addrbuf, &sotoinpcb(head)->inp_inc.inc6_laddr); sbuf_printf(&descrsb, "[%s]", addrbuf); } else #endif { #ifdef INET inet_ntoa_r( sotoinpcb(head)->inp_inc.inc_laddr, addrbuf); sbuf_cat(&descrsb, addrbuf); #endif } sbuf_printf(&descrsb, ":%hu (proto %u)", ntohs(sotoinpcb(head)->inp_inc.inc_lport), head->so_proto->pr_protocol); break; #endif /* INET || INET6 */ case AF_UNIX: sbuf_cat(&descrsb, localprefix); if (sotounpcb(head)->unp_addr != NULL) len = sotounpcb(head)->unp_addr->sun_len - offsetof(struct sockaddr_un, sun_path); else len = 0; if (len > 0) sbuf_bcat(&descrsb, sotounpcb(head)->unp_addr->sun_path, len); else sbuf_cat(&descrsb, "(unknown)"); break; } /* * If we can't print something more specific, at least * print the domain name. */ if (sbuf_finish(&descrsb) != 0 || sbuf_len(&descrsb) <= 0) { sbuf_clear(&descrsb); sbuf_cat(&descrsb, head->so_proto->pr_domain->dom_name ?: "unknown"); sbuf_finish(&descrsb); } KASSERT(sbuf_len(&descrsb) > 0, ("%s: sbuf creation failed", __func__)); log(LOG_DEBUG, "%s: pcb %p (%s): Listen queue overflow: " "%i already in queue awaiting acceptance " "(%d occurrences)\n", __func__, head->so_pcb, sbuf_data(&descrsb), qlen, overcount); sbuf_delete(&descrsb); overcount = 0; } return (NULL); } SOLISTEN_UNLOCK(head); VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", __func__, head)); so = soalloc(head->so_vnet); if (so == NULL) { log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " "limit reached or out of memory\n", __func__, head->so_pcb); return (NULL); } so->so_listen = head; so->so_type = head->so_type; so->so_linger = head->so_linger; so->so_state = head->so_state | SS_NOFDREF; so->so_fibnum = head->so_fibnum; so->so_proto = head->so_proto; so->so_cred = crhold(head->so_cred); #ifdef MAC mac_socket_newconn(head, so); #endif knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, - so_rdknl_assert_locked, so_rdknl_assert_unlocked); + so_rdknl_assert_lock); knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, - so_wrknl_assert_locked, so_wrknl_assert_unlocked); + so_wrknl_assert_lock); VNET_SO_ASSERT(head); if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { sodealloc(so); log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", __func__, head->so_pcb); return (NULL); } if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { sodealloc(so); log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", __func__, head->so_pcb); return (NULL); } so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; so->so_snd.sb_lowat = head->sol_sbsnd_lowat; so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; so->so_snd.sb_timeo = head->sol_sbsnd_timeo; so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE; so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE; SOLISTEN_LOCK(head); if (head->sol_accept_filter != NULL) connstatus = 0; so->so_state |= connstatus; so->so_options = head->so_options & ~SO_ACCEPTCONN; soref(head); /* A socket on (in)complete queue refs head. */ if (connstatus) { TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); so->so_qstate = SQ_COMP; head->sol_qlen++; solisten_wakeup(head); /* unlocks */ } else { /* * Keep removing sockets from the head until there's room for * us to insert on the tail. In pre-locking revisions, this * was a simple if(), but as we could be racing with other * threads and soabort() requires dropping locks, we must * loop waiting for the condition to be true. */ while (head->sol_incqlen > head->sol_qlimit) { struct socket *sp; sp = TAILQ_FIRST(&head->sol_incomp); TAILQ_REMOVE(&head->sol_incomp, sp, so_list); head->sol_incqlen--; SOCK_LOCK(sp); sp->so_qstate = SQ_NONE; sp->so_listen = NULL; SOCK_UNLOCK(sp); sorele(head); /* does SOLISTEN_UNLOCK, head stays */ soabort(sp); SOLISTEN_LOCK(head); } TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); so->so_qstate = SQ_INCOMP; head->sol_incqlen++; SOLISTEN_UNLOCK(head); } return (so); } #if defined(SCTP) || defined(SCTP_SUPPORT) /* * Socket part of sctp_peeloff(). Detach a new socket from an * association. The new socket is returned with a reference. */ struct socket * sopeeloff(struct socket *head) { struct socket *so; VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", __func__, __LINE__, head)); so = soalloc(head->so_vnet); if (so == NULL) { log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " "limit reached or out of memory\n", __func__, head->so_pcb); return (NULL); } so->so_type = head->so_type; so->so_options = head->so_options; so->so_linger = head->so_linger; so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; so->so_fibnum = head->so_fibnum; so->so_proto = head->so_proto; so->so_cred = crhold(head->so_cred); #ifdef MAC mac_socket_newconn(head, so); #endif knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, - so_rdknl_assert_locked, so_rdknl_assert_unlocked); + so_rdknl_assert_lock); knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, - so_wrknl_assert_locked, so_wrknl_assert_unlocked); + so_wrknl_assert_lock); VNET_SO_ASSERT(head); if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { sodealloc(so); log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", __func__, head->so_pcb); return (NULL); } if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { sodealloc(so); log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", __func__, head->so_pcb); return (NULL); } so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; so->so_snd.sb_lowat = head->so_snd.sb_lowat; so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; so->so_snd.sb_timeo = head->so_snd.sb_timeo; so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; soref(so); return (so); } #endif /* SCTP */ int sobind(struct socket *so, struct sockaddr *nam, struct thread *td) { int error; CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); CURVNET_RESTORE(); return (error); } int sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { int error; CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); CURVNET_RESTORE(); return (error); } /* * solisten() transitions a socket from a non-listening state to a listening * state, but can also be used to update the listen queue depth on an * existing listen socket. The protocol will call back into the sockets * layer using solisten_proto_check() and solisten_proto() to check and set * socket-layer listen state. Call backs are used so that the protocol can * acquire both protocol and socket layer locks in whatever order is required * by the protocol. * * Protocol implementors are advised to hold the socket lock across the * socket-layer test and set to avoid races at the socket layer. */ int solisten(struct socket *so, int backlog, struct thread *td) { int error; CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); CURVNET_RESTORE(); return (error); } int solisten_proto_check(struct socket *so) { SOCK_LOCK_ASSERT(so); if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) return (EINVAL); return (0); } void solisten_proto(struct socket *so, int backlog) { int sbrcv_lowat, sbsnd_lowat; u_int sbrcv_hiwat, sbsnd_hiwat; short sbrcv_flags, sbsnd_flags; sbintime_t sbrcv_timeo, sbsnd_timeo; SOCK_LOCK_ASSERT(so); if (SOLISTENING(so)) goto listening; /* * Change this socket to listening state. */ sbrcv_lowat = so->so_rcv.sb_lowat; sbsnd_lowat = so->so_snd.sb_lowat; sbrcv_hiwat = so->so_rcv.sb_hiwat; sbsnd_hiwat = so->so_snd.sb_hiwat; sbrcv_flags = so->so_rcv.sb_flags; sbsnd_flags = so->so_snd.sb_flags; sbrcv_timeo = so->so_rcv.sb_timeo; sbsnd_timeo = so->so_snd.sb_timeo; sbdestroy(&so->so_snd, so); sbdestroy(&so->so_rcv, so); sx_destroy(&so->so_snd.sb_sx); sx_destroy(&so->so_rcv.sb_sx); SOCKBUF_LOCK_DESTROY(&so->so_snd); SOCKBUF_LOCK_DESTROY(&so->so_rcv); #ifdef INVARIANTS bzero(&so->so_rcv, sizeof(struct socket) - offsetof(struct socket, so_rcv)); #endif so->sol_sbrcv_lowat = sbrcv_lowat; so->sol_sbsnd_lowat = sbsnd_lowat; so->sol_sbrcv_hiwat = sbrcv_hiwat; so->sol_sbsnd_hiwat = sbsnd_hiwat; so->sol_sbrcv_flags = sbrcv_flags; so->sol_sbsnd_flags = sbsnd_flags; so->sol_sbrcv_timeo = sbrcv_timeo; so->sol_sbsnd_timeo = sbsnd_timeo; so->sol_qlen = so->sol_incqlen = 0; TAILQ_INIT(&so->sol_incomp); TAILQ_INIT(&so->sol_comp); so->sol_accept_filter = NULL; so->sol_accept_filter_arg = NULL; so->sol_accept_filter_str = NULL; so->sol_upcall = NULL; so->sol_upcallarg = NULL; so->so_options |= SO_ACCEPTCONN; listening: if (backlog < 0 || backlog > somaxconn) backlog = somaxconn; so->sol_qlimit = backlog; } /* * Wakeup listeners/subsystems once we have a complete connection. * Enters with lock, returns unlocked. */ void solisten_wakeup(struct socket *sol) { if (sol->sol_upcall != NULL) (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); else { selwakeuppri(&sol->so_rdsel, PSOCK); KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); } SOLISTEN_UNLOCK(sol); wakeup_one(&sol->sol_comp); if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) pgsigio(&sol->so_sigio, SIGIO, 0); } /* * Return single connection off a listening socket queue. Main consumer of * the function is kern_accept4(). Some modules, that do their own accept * management also use the function. * * Listening socket must be locked on entry and is returned unlocked on * return. * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. */ int solisten_dequeue(struct socket *head, struct socket **ret, int flags) { struct socket *so; int error; SOLISTEN_LOCK_ASSERT(head); while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && head->so_error == 0) { error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH, "accept", 0); if (error != 0) { SOLISTEN_UNLOCK(head); return (error); } } if (head->so_error) { error = head->so_error; head->so_error = 0; } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) error = EWOULDBLOCK; else error = 0; if (error) { SOLISTEN_UNLOCK(head); return (error); } so = TAILQ_FIRST(&head->sol_comp); SOCK_LOCK(so); KASSERT(so->so_qstate == SQ_COMP, ("%s: so %p not SQ_COMP", __func__, so)); soref(so); head->sol_qlen--; so->so_qstate = SQ_NONE; so->so_listen = NULL; TAILQ_REMOVE(&head->sol_comp, so, so_list); if (flags & ACCEPT4_INHERIT) so->so_state |= (head->so_state & SS_NBIO); else so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; SOCK_UNLOCK(so); sorele(head); *ret = so; return (0); } /* * Evaluate the reference count and named references on a socket; if no * references remain, free it. This should be called whenever a reference is * released, such as in sorele(), but also when named reference flags are * cleared in socket or protocol code. * * sofree() will free the socket if: * * - There are no outstanding file descriptor references or related consumers * (so_count == 0). * * - The socket has been closed by user space, if ever open (SS_NOFDREF). * * - The protocol does not have an outstanding strong reference on the socket * (SS_PROTOREF). * * - The socket is not in a completed connection queue, so a process has been * notified that it is present. If it is removed, the user process may * block in accept() despite select() saying the socket was ready. */ void sofree(struct socket *so) { struct protosw *pr = so->so_proto; SOCK_LOCK_ASSERT(so); if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) { SOCK_UNLOCK(so); return; } if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) { struct socket *sol; sol = so->so_listen; KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so)); /* * To solve race between close of a listening socket and * a socket on its incomplete queue, we need to lock both. * The order is first listening socket, then regular. * Since we don't have SS_NOFDREF neither SS_PROTOREF, this * function and the listening socket are the only pointers * to so. To preserve so and sol, we reference both and then * relock. * After relock the socket may not move to so_comp since it * doesn't have PCB already, but it may be removed from * so_incomp. If that happens, we share responsiblity on * freeing the socket, but soclose() has already removed * it from queue. */ soref(sol); soref(so); SOCK_UNLOCK(so); SOLISTEN_LOCK(sol); SOCK_LOCK(so); if (so->so_qstate == SQ_INCOMP) { KASSERT(so->so_listen == sol, ("%s: so %p migrated out of sol %p", __func__, so, sol)); TAILQ_REMOVE(&sol->sol_incomp, so, so_list); sol->sol_incqlen--; /* This is guarenteed not to be the last. */ refcount_release(&sol->so_count); so->so_qstate = SQ_NONE; so->so_listen = NULL; } else KASSERT(so->so_listen == NULL, ("%s: so %p not on (in)comp with so_listen", __func__, so)); sorele(sol); KASSERT(so->so_count == 1, ("%s: so %p count %u", __func__, so, so->so_count)); so->so_count = 0; } if (SOLISTENING(so)) so->so_error = ECONNABORTED; SOCK_UNLOCK(so); if (so->so_dtor != NULL) so->so_dtor(so); VNET_SO_ASSERT(so); if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) (*pr->pr_domain->dom_dispose)(so); if (pr->pr_usrreqs->pru_detach != NULL) (*pr->pr_usrreqs->pru_detach)(so); /* * From this point on, we assume that no other references to this * socket exist anywhere else in the stack. Therefore, no locks need * to be acquired or held. * * We used to do a lot of socket buffer and socket locking here, as * well as invoke sorflush() and perform wakeups. The direct call to * dom_dispose() and sbdestroy() are an inlining of what was * necessary from sorflush(). * * Notice that the socket buffer and kqueue state are torn down * before calling pru_detach. This means that protocols shold not * assume they can perform socket wakeups, etc, in their detach code. */ if (!SOLISTENING(so)) { sbdestroy(&so->so_snd, so); sbdestroy(&so->so_rcv, so); } seldrain(&so->so_rdsel); seldrain(&so->so_wrsel); knlist_destroy(&so->so_rdsel.si_note); knlist_destroy(&so->so_wrsel.si_note); sodealloc(so); } /* * Close a socket on last file table reference removal. Initiate disconnect * if connected. Free socket when disconnect complete. * * This function will sorele() the socket. Note that soclose() may be called * prior to the ref count reaching zero. The actual socket structure will * not be freed until the ref count reaches zero. */ int soclose(struct socket *so) { struct accept_queue lqueue; bool listening; int error = 0; KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); CURVNET_SET(so->so_vnet); funsetown(&so->so_sigio); if (so->so_state & SS_ISCONNECTED) { if ((so->so_state & SS_ISDISCONNECTING) == 0) { error = sodisconnect(so); if (error) { if (error == ENOTCONN) error = 0; goto drop; } } if (so->so_options & SO_LINGER) { if ((so->so_state & SS_ISDISCONNECTING) && (so->so_state & SS_NBIO)) goto drop; while (so->so_state & SS_ISCONNECTED) { error = tsleep(&so->so_timeo, PSOCK | PCATCH, "soclos", so->so_linger * hz); if (error) break; } } } drop: if (so->so_proto->pr_usrreqs->pru_close != NULL) (*so->so_proto->pr_usrreqs->pru_close)(so); SOCK_LOCK(so); if ((listening = (so->so_options & SO_ACCEPTCONN))) { struct socket *sp; TAILQ_INIT(&lqueue); TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); so->sol_qlen = so->sol_incqlen = 0; TAILQ_FOREACH(sp, &lqueue, so_list) { SOCK_LOCK(sp); sp->so_qstate = SQ_NONE; sp->so_listen = NULL; SOCK_UNLOCK(sp); /* Guaranteed not to be the last. */ refcount_release(&so->so_count); } } KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); so->so_state |= SS_NOFDREF; sorele(so); if (listening) { struct socket *sp, *tsp; TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) { SOCK_LOCK(sp); if (sp->so_count == 0) { SOCK_UNLOCK(sp); soabort(sp); } else /* sp is now in sofree() */ SOCK_UNLOCK(sp); } } CURVNET_RESTORE(); return (error); } /* * soabort() is used to abruptly tear down a connection, such as when a * resource limit is reached (listen queue depth exceeded), or if a listen * socket is closed while there are sockets waiting to be accepted. * * This interface is tricky, because it is called on an unreferenced socket, * and must be called only by a thread that has actually removed the socket * from the listen queue it was on, or races with other threads are risked. * * This interface will call into the protocol code, so must not be called * with any socket locks held. Protocols do call it while holding their own * recursible protocol mutexes, but this is something that should be subject * to review in the future. */ void soabort(struct socket *so) { /* * In as much as is possible, assert that no references to this * socket are held. This is not quite the same as asserting that the * current thread is responsible for arranging for no references, but * is as close as we can get for now. */ KASSERT(so->so_count == 0, ("soabort: so_count")); KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); VNET_SO_ASSERT(so); if (so->so_proto->pr_usrreqs->pru_abort != NULL) (*so->so_proto->pr_usrreqs->pru_abort)(so); SOCK_LOCK(so); sofree(so); } int soaccept(struct socket *so, struct sockaddr **nam) { int error; SOCK_LOCK(so); KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); so->so_state &= ~SS_NOFDREF; SOCK_UNLOCK(so); CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); CURVNET_RESTORE(); return (error); } int soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) { return (soconnectat(AT_FDCWD, so, nam, td)); } int soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { int error; if (so->so_options & SO_ACCEPTCONN) return (EOPNOTSUPP); CURVNET_SET(so->so_vnet); /* * If protocol is connection-based, can only connect once. * Otherwise, if connected, try to disconnect first. This allows * user to disconnect by connecting to, e.g., a null address. */ if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && ((so->so_proto->pr_flags & PR_CONNREQUIRED) || (error = sodisconnect(so)))) { error = EISCONN; } else { /* * Prevent accumulated error from previous connection from * biting us. */ so->so_error = 0; if (fd == AT_FDCWD) { error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td); } else { error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, so, nam, td); } } CURVNET_RESTORE(); return (error); } int soconnect2(struct socket *so1, struct socket *so2) { int error; CURVNET_SET(so1->so_vnet); error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); CURVNET_RESTORE(); return (error); } int sodisconnect(struct socket *so) { int error; if ((so->so_state & SS_ISCONNECTED) == 0) return (ENOTCONN); if (so->so_state & SS_ISDISCONNECTING) return (EALREADY); VNET_SO_ASSERT(so); error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); return (error); } #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) int sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { long space; ssize_t resid; int clen = 0, error, dontroute; KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); KASSERT(so->so_proto->pr_flags & PR_ATOMIC, ("sosend_dgram: !PR_ATOMIC")); if (uio != NULL) resid = uio->uio_resid; else resid = top->m_pkthdr.len; /* * In theory resid should be unsigned. However, space must be * signed, as it might be less than 0 if we over-committed, and we * must use a signed comparison of space and resid. On the other * hand, a negative resid causes us to loop sending 0-length * segments to the protocol. */ if (resid < 0) { error = EINVAL; goto out; } dontroute = (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; if (td != NULL) td->td_ru.ru_msgsnd++; if (control != NULL) clen = control->m_len; SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_state & SBS_CANTSENDMORE) { SOCKBUF_UNLOCK(&so->so_snd); error = EPIPE; goto out; } if (so->so_error) { error = so->so_error; so->so_error = 0; SOCKBUF_UNLOCK(&so->so_snd); goto out; } if ((so->so_state & SS_ISCONNECTED) == 0) { /* * `sendto' and `sendmsg' is allowed on a connection-based * socket if it supports implied connect. Return ENOTCONN if * not connected and no address is supplied. */ if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { if ((so->so_state & SS_ISCONFIRMING) == 0 && !(resid == 0 && clen != 0)) { SOCKBUF_UNLOCK(&so->so_snd); error = ENOTCONN; goto out; } } else if (addr == NULL) { if (so->so_proto->pr_flags & PR_CONNREQUIRED) error = ENOTCONN; else error = EDESTADDRREQ; SOCKBUF_UNLOCK(&so->so_snd); goto out; } } /* * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a * problem and need fixing. */ space = sbspace(&so->so_snd); if (flags & MSG_OOB) space += 1024; space -= clen; SOCKBUF_UNLOCK(&so->so_snd); if (resid > space) { error = EMSGSIZE; goto out; } if (uio == NULL) { resid = 0; if (flags & MSG_EOR) top->m_flags |= M_EOR; } else { /* * Copy the data from userland into a mbuf chain. * If no data is to be copied in, a single empty mbuf * is returned. */ top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); if (top == NULL) { error = EFAULT; /* only possible error */ goto out; } space -= resid - uio->uio_resid; resid = uio->uio_resid; } KASSERT(resid == 0, ("sosend_dgram: resid != 0")); /* * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock * than with. */ if (dontroute) { SOCK_LOCK(so); so->so_options |= SO_DONTROUTE; SOCK_UNLOCK(so); } /* * XXX all the SBS_CANTSENDMORE checks previously done could be out * of date. We could have received a reset packet in an interrupt or * maybe we slept while doing page faults in uiomove() etc. We could * probably recheck again inside the locking protection here, but * there are probably other places that this also happens. We must * rethink this. */ VNET_SO_ASSERT(so); error = (*so->so_proto->pr_usrreqs->pru_send)(so, (flags & MSG_OOB) ? PRUS_OOB : /* * If the user set MSG_EOF, the protocol understands this flag and * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. */ ((flags & MSG_EOF) && (so->so_proto->pr_flags & PR_IMPLOPCL) && (resid <= 0)) ? PRUS_EOF : /* If there is more to send set PRUS_MORETOCOME */ (flags & MSG_MORETOCOME) || (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, top, addr, control, td); if (dontroute) { SOCK_LOCK(so); so->so_options &= ~SO_DONTROUTE; SOCK_UNLOCK(so); } clen = 0; control = NULL; top = NULL; out: if (top != NULL) m_freem(top); if (control != NULL) m_freem(control); return (error); } /* * Send on a socket. If send must go all at once and message is larger than * send buffering, then hard error. Lock against other senders. If must go * all at once and not enough room now, then inform user that this would * block and do nothing. Otherwise, if nonblocking, send as much as * possible. The data to be sent is described by "uio" if nonzero, otherwise * by the mbuf chain "top" (which must be null if uio is not). Data provided * in mbuf chain must be small enough to send all at once. * * Returns nonzero on error, timeout or signal; callers must check for short * counts if EINTR/ERESTART are returned. Data and control buffers are freed * on return. */ int sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { long space; ssize_t resid; int clen = 0, error, dontroute; int atomic = sosendallatonce(so) || top; int pru_flag; #ifdef KERN_TLS struct ktls_session *tls; int tls_enq_cnt, tls_pruflag; uint8_t tls_rtype; tls = NULL; tls_rtype = TLS_RLTYPE_APP; #endif if (uio != NULL) resid = uio->uio_resid; else if ((top->m_flags & M_PKTHDR) != 0) resid = top->m_pkthdr.len; else resid = m_length(top, NULL); /* * In theory resid should be unsigned. However, space must be * signed, as it might be less than 0 if we over-committed, and we * must use a signed comparison of space and resid. On the other * hand, a negative resid causes us to loop sending 0-length * segments to the protocol. * * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM * type sockets since that's an error. */ if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { error = EINVAL; goto out; } dontroute = (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && (so->so_proto->pr_flags & PR_ATOMIC); if (td != NULL) td->td_ru.ru_msgsnd++; if (control != NULL) clen = control->m_len; error = sblock(&so->so_snd, SBLOCKWAIT(flags)); if (error) goto out; #ifdef KERN_TLS tls_pruflag = 0; tls = ktls_hold(so->so_snd.sb_tls_info); if (tls != NULL) { if (tls->mode == TCP_TLS_MODE_SW) tls_pruflag = PRUS_NOTREADY; if (control != NULL) { struct cmsghdr *cm = mtod(control, struct cmsghdr *); if (clen >= sizeof(*cm) && cm->cmsg_type == TLS_SET_RECORD_TYPE) { tls_rtype = *((uint8_t *)CMSG_DATA(cm)); clen = 0; m_freem(control); control = NULL; atomic = 1; } } } #endif restart: do { SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_state & SBS_CANTSENDMORE) { SOCKBUF_UNLOCK(&so->so_snd); error = EPIPE; goto release; } if (so->so_error) { error = so->so_error; so->so_error = 0; SOCKBUF_UNLOCK(&so->so_snd); goto release; } if ((so->so_state & SS_ISCONNECTED) == 0) { /* * `sendto' and `sendmsg' is allowed on a connection- * based socket if it supports implied connect. * Return ENOTCONN if not connected and no address is * supplied. */ if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { if ((so->so_state & SS_ISCONFIRMING) == 0 && !(resid == 0 && clen != 0)) { SOCKBUF_UNLOCK(&so->so_snd); error = ENOTCONN; goto release; } } else if (addr == NULL) { SOCKBUF_UNLOCK(&so->so_snd); if (so->so_proto->pr_flags & PR_CONNREQUIRED) error = ENOTCONN; else error = EDESTADDRREQ; goto release; } } space = sbspace(&so->so_snd); if (flags & MSG_OOB) space += 1024; if ((atomic && resid > so->so_snd.sb_hiwat) || clen > so->so_snd.sb_hiwat) { SOCKBUF_UNLOCK(&so->so_snd); error = EMSGSIZE; goto release; } if (space < resid + clen && (atomic || space < so->so_snd.sb_lowat || space < clen)) { if ((so->so_state & SS_NBIO) || (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { SOCKBUF_UNLOCK(&so->so_snd); error = EWOULDBLOCK; goto release; } error = sbwait(&so->so_snd); SOCKBUF_UNLOCK(&so->so_snd); if (error) goto release; goto restart; } SOCKBUF_UNLOCK(&so->so_snd); space -= clen; do { if (uio == NULL) { resid = 0; if (flags & MSG_EOR) top->m_flags |= M_EOR; #ifdef KERN_TLS if (tls != NULL) { ktls_frame(top, tls, &tls_enq_cnt, tls_rtype); tls_rtype = TLS_RLTYPE_APP; } #endif } else { /* * Copy the data from userland into a mbuf * chain. If resid is 0, which can happen * only if we have control to send, then * a single empty mbuf is returned. This * is a workaround to prevent protocol send * methods to panic. */ #ifdef KERN_TLS if (tls != NULL) { top = m_uiotombuf(uio, M_WAITOK, space, tls->params.max_frame_len, M_EXTPG | ((flags & MSG_EOR) ? M_EOR : 0)); if (top != NULL) { ktls_frame(top, tls, &tls_enq_cnt, tls_rtype); } tls_rtype = TLS_RLTYPE_APP; } else #endif top = m_uiotombuf(uio, M_WAITOK, space, (atomic ? max_hdr : 0), (atomic ? M_PKTHDR : 0) | ((flags & MSG_EOR) ? M_EOR : 0)); if (top == NULL) { error = EFAULT; /* only possible error */ goto release; } space -= resid - uio->uio_resid; resid = uio->uio_resid; } if (dontroute) { SOCK_LOCK(so); so->so_options |= SO_DONTROUTE; SOCK_UNLOCK(so); } /* * XXX all the SBS_CANTSENDMORE checks previously * done could be out of date. We could have received * a reset packet in an interrupt or maybe we slept * while doing page faults in uiomove() etc. We * could probably recheck again inside the locking * protection here, but there are probably other * places that this also happens. We must rethink * this. */ VNET_SO_ASSERT(so); pru_flag = (flags & MSG_OOB) ? PRUS_OOB : /* * If the user set MSG_EOF, the protocol understands * this flag and nothing left to send then use * PRU_SEND_EOF instead of PRU_SEND. */ ((flags & MSG_EOF) && (so->so_proto->pr_flags & PR_IMPLOPCL) && (resid <= 0)) ? PRUS_EOF : /* If there is more to send set PRUS_MORETOCOME. */ (flags & MSG_MORETOCOME) || (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; #ifdef KERN_TLS pru_flag |= tls_pruflag; #endif error = (*so->so_proto->pr_usrreqs->pru_send)(so, pru_flag, top, addr, control, td); if (dontroute) { SOCK_LOCK(so); so->so_options &= ~SO_DONTROUTE; SOCK_UNLOCK(so); } #ifdef KERN_TLS if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { /* * Note that error is intentionally * ignored. * * Like sendfile(), we rely on the * completion routine (pru_ready()) * to free the mbufs in the event that * pru_send() encountered an error and * did not append them to the sockbuf. */ soref(so); ktls_enqueue(top, so, tls_enq_cnt); } #endif clen = 0; control = NULL; top = NULL; if (error) goto release; } while (resid && space > 0); } while (resid); release: sbunlock(&so->so_snd); out: #ifdef KERN_TLS if (tls != NULL) ktls_free(tls); #endif if (top != NULL) m_freem(top); if (control != NULL) m_freem(control); return (error); } int sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { int error; CURVNET_SET(so->so_vnet); if (!SOLISTENING(so)) error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top, control, flags, td); else { m_freem(top); m_freem(control); error = ENOTCONN; } CURVNET_RESTORE(); return (error); } /* * The part of soreceive() that implements reading non-inline out-of-band * data from a socket. For more complete comments, see soreceive(), from * which this code originated. * * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is * unable to return an mbuf chain to the caller. */ static int soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) { struct protosw *pr = so->so_proto; struct mbuf *m; int error; KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); VNET_SO_ASSERT(so); m = m_get(M_WAITOK, MT_DATA); error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); if (error) goto bad; do { error = uiomove(mtod(m, void *), (int) min(uio->uio_resid, m->m_len), uio); m = m_free(m); } while (uio->uio_resid && error == 0 && m); bad: if (m != NULL) m_freem(m); return (error); } /* * Following replacement or removal of the first mbuf on the first mbuf chain * of a socket buffer, push necessary state changes back into the socket * buffer so that other consumers see the values consistently. 'nextrecord' * is the callers locally stored value of the original value of * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. * NOTE: 'nextrecord' may be NULL. */ static __inline void sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) { SOCKBUF_LOCK_ASSERT(sb); /* * First, update for the new value of nextrecord. If necessary, make * it the first record. */ if (sb->sb_mb != NULL) sb->sb_mb->m_nextpkt = nextrecord; else sb->sb_mb = nextrecord; /* * Now update any dependent socket buffer fields to reflect the new * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the * addition of a second clause that takes care of the case where * sb_mb has been updated, but remains the last record. */ if (sb->sb_mb == NULL) { sb->sb_mbtail = NULL; sb->sb_lastrecord = NULL; } else if (sb->sb_mb->m_nextpkt == NULL) sb->sb_lastrecord = sb->sb_mb; } /* * Implement receive operations on a socket. We depend on the way that * records are added to the sockbuf by sbappend. In particular, each record * (mbufs linked through m_next) must begin with an address if the protocol * so specifies, followed by an optional mbuf or mbufs containing ancillary * data, and then zero or more mbufs of data. In order to allow parallelism * between network receive and copying to user space, as well as avoid * sleeping with a mutex held, we release the socket buffer mutex during the * user space copy. Although the sockbuf is locked, new data may still be * appended, and thus we must maintain consistency of the sockbuf during that * time. * * The caller may receive the data as a single mbuf chain by supplying an * mbuf **mp0 for use in returning the chain. The uio is then used only for * the count in uio_resid. */ int soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { struct mbuf *m, **mp; int flags, error, offset; ssize_t len; struct protosw *pr = so->so_proto; struct mbuf *nextrecord; int moff, type = 0; ssize_t orig_resid = uio->uio_resid; mp = mp0; if (psa != NULL) *psa = NULL; if (controlp != NULL) *controlp = NULL; if (flagsp != NULL) flags = *flagsp &~ MSG_EOR; else flags = 0; if (flags & MSG_OOB) return (soreceive_rcvoob(so, uio, flags)); if (mp != NULL) *mp = NULL; if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) && uio->uio_resid) { VNET_SO_ASSERT(so); (*pr->pr_usrreqs->pru_rcvd)(so, 0); } error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); if (error) return (error); restart: SOCKBUF_LOCK(&so->so_rcv); m = so->so_rcv.sb_mb; /* * If we have less data than requested, block awaiting more (subject * to any timeout) if: * 1. the current count is less than the low water mark, or * 2. MSG_DONTWAIT is not set */ if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && sbavail(&so->so_rcv) < uio->uio_resid) && sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { KASSERT(m != NULL || !sbavail(&so->so_rcv), ("receive: m == %p sbavail == %u", m, sbavail(&so->so_rcv))); if (so->so_error) { if (m != NULL) goto dontblock; error = so->so_error; if ((flags & MSG_PEEK) == 0) so->so_error = 0; SOCKBUF_UNLOCK(&so->so_rcv); goto release; } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { if (m != NULL) goto dontblock; #ifdef KERN_TLS else if (so->so_rcv.sb_tlsdcc == 0 && so->so_rcv.sb_tlscc == 0) { #else else { #endif SOCKBUF_UNLOCK(&so->so_rcv); goto release; } } for (; m != NULL; m = m->m_next) if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { m = so->so_rcv.sb_mb; goto dontblock; } if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { SOCKBUF_UNLOCK(&so->so_rcv); error = ENOTCONN; goto release; } if (uio->uio_resid == 0) { SOCKBUF_UNLOCK(&so->so_rcv); goto release; } if ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO))) { SOCKBUF_UNLOCK(&so->so_rcv); error = EWOULDBLOCK; goto release; } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); error = sbwait(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); if (error) goto release; goto restart; } dontblock: /* * From this point onward, we maintain 'nextrecord' as a cache of the * pointer to the next record in the socket buffer. We must keep the * various socket buffer pointers and local stack versions of the * pointers in sync, pushing out modifications before dropping the * socket buffer mutex, and re-reading them when picking it up. * * Otherwise, we will race with the network stack appending new data * or records onto the socket buffer by using inconsistent/stale * versions of the field, possibly resulting in socket buffer * corruption. * * By holding the high-level sblock(), we prevent simultaneous * readers from pulling off the front of the socket buffer. */ SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (uio->uio_td) uio->uio_td->td_ru.ru_msgrcv++; KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); nextrecord = m->m_nextpkt; if (pr->pr_flags & PR_ADDR) { KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); orig_resid = 0; if (psa != NULL) *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_NOWAIT); if (flags & MSG_PEEK) { m = m->m_next; } else { sbfree(&so->so_rcv, m); so->so_rcv.sb_mb = m_free(m); m = so->so_rcv.sb_mb; sockbuf_pushsync(&so->so_rcv, nextrecord); } } /* * Process one or more MT_CONTROL mbufs present before any data mbufs * in the first mbuf chain on the socket buffer. If MSG_PEEK, we * just copy the data; if !MSG_PEEK, we call into the protocol to * perform externalization (or freeing if controlp == NULL). */ if (m != NULL && m->m_type == MT_CONTROL) { struct mbuf *cm = NULL, *cmn; struct mbuf **cme = &cm; #ifdef KERN_TLS struct cmsghdr *cmsg; struct tls_get_record tgr; /* * For MSG_TLSAPPDATA, check for a non-application data * record. If found, return ENXIO without removing * it from the receive queue. This allows a subsequent * call without MSG_TLSAPPDATA to receive it. * Note that, for TLS, there should only be a single * control mbuf with the TLS_GET_RECORD message in it. */ if (flags & MSG_TLSAPPDATA) { cmsg = mtod(m, struct cmsghdr *); if (cmsg->cmsg_type == TLS_GET_RECORD && cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); /* This will need to change for TLS 1.3. */ if (tgr.tls_type != TLS_RLTYPE_APP) { SOCKBUF_UNLOCK(&so->so_rcv); error = ENXIO; goto release; } } } #endif do { if (flags & MSG_PEEK) { if (controlp != NULL) { *controlp = m_copym(m, 0, m->m_len, M_NOWAIT); controlp = &(*controlp)->m_next; } m = m->m_next; } else { sbfree(&so->so_rcv, m); so->so_rcv.sb_mb = m->m_next; m->m_next = NULL; *cme = m; cme = &(*cme)->m_next; m = so->so_rcv.sb_mb; } } while (m != NULL && m->m_type == MT_CONTROL); if ((flags & MSG_PEEK) == 0) sockbuf_pushsync(&so->so_rcv, nextrecord); while (cm != NULL) { cmn = cm->m_next; cm->m_next = NULL; if (pr->pr_domain->dom_externalize != NULL) { SOCKBUF_UNLOCK(&so->so_rcv); VNET_SO_ASSERT(so); error = (*pr->pr_domain->dom_externalize) (cm, controlp, flags); SOCKBUF_LOCK(&so->so_rcv); } else if (controlp != NULL) *controlp = cm; else m_freem(cm); if (controlp != NULL) { orig_resid = 0; while (*controlp != NULL) controlp = &(*controlp)->m_next; } cm = cmn; } if (m != NULL) nextrecord = so->so_rcv.sb_mb->m_nextpkt; else nextrecord = so->so_rcv.sb_mb; orig_resid = 0; } if (m != NULL) { if ((flags & MSG_PEEK) == 0) { KASSERT(m->m_nextpkt == nextrecord, ("soreceive: post-control, nextrecord !sync")); if (nextrecord == NULL) { KASSERT(so->so_rcv.sb_mb == m, ("soreceive: post-control, sb_mb!=m")); KASSERT(so->so_rcv.sb_lastrecord == m, ("soreceive: post-control, lastrecord!=m")); } } type = m->m_type; if (type == MT_OOBDATA) flags |= MSG_OOB; } else { if ((flags & MSG_PEEK) == 0) { KASSERT(so->so_rcv.sb_mb == nextrecord, ("soreceive: sb_mb != nextrecord")); if (so->so_rcv.sb_mb == NULL) { KASSERT(so->so_rcv.sb_lastrecord == NULL, ("soreceive: sb_lastercord != NULL")); } } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); /* * Now continue to read any data mbufs off of the head of the socket * buffer until the read request is satisfied. Note that 'type' is * used to store the type of any mbuf reads that have happened so far * such that soreceive() can stop reading if the type changes, which * causes soreceive() to return only one of regular data and inline * out-of-band data in a single socket receive operation. */ moff = 0; offset = 0; while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 && error == 0) { /* * If the type of mbuf has changed since the last mbuf * examined ('type'), end the receive operation. */ SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { if (type != m->m_type) break; } else if (type == MT_OOBDATA) break; else KASSERT(m->m_type == MT_DATA, ("m->m_type == %d", m->m_type)); so->so_rcv.sb_state &= ~SBS_RCVATMARK; len = uio->uio_resid; if (so->so_oobmark && len > so->so_oobmark - offset) len = so->so_oobmark - offset; if (len > m->m_len - moff) len = m->m_len - moff; /* * If mp is set, just pass back the mbufs. Otherwise copy * them out via the uio, then free. Sockbuf must be * consistent here (points to current mbuf, it points to next * record) when we drop priority; we must note any additions * to the sockbuf when we block interrupts again. */ if (mp == NULL) { SOCKBUF_LOCK_ASSERT(&so->so_rcv); SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); if ((m->m_flags & M_EXTPG) != 0) error = m_unmappedtouio(m, moff, uio, (int)len); else error = uiomove(mtod(m, char *) + moff, (int)len, uio); SOCKBUF_LOCK(&so->so_rcv); if (error) { /* * The MT_SONAME mbuf has already been removed * from the record, so it is necessary to * remove the data mbufs, if any, to preserve * the invariant in the case of PR_ADDR that * requires MT_SONAME mbufs at the head of * each record. */ if (pr->pr_flags & PR_ATOMIC && ((flags & MSG_PEEK) == 0)) (void)sbdroprecord_locked(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); goto release; } } else uio->uio_resid -= len; SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (len == m->m_len - moff) { if (m->m_flags & M_EOR) flags |= MSG_EOR; if (flags & MSG_PEEK) { m = m->m_next; moff = 0; } else { nextrecord = m->m_nextpkt; sbfree(&so->so_rcv, m); if (mp != NULL) { m->m_nextpkt = NULL; *mp = m; mp = &m->m_next; so->so_rcv.sb_mb = m = m->m_next; *mp = NULL; } else { so->so_rcv.sb_mb = m_free(m); m = so->so_rcv.sb_mb; } sockbuf_pushsync(&so->so_rcv, nextrecord); SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); } } else { if (flags & MSG_PEEK) moff += len; else { if (mp != NULL) { if (flags & MSG_DONTWAIT) { *mp = m_copym(m, 0, len, M_NOWAIT); if (*mp == NULL) { /* * m_copym() couldn't * allocate an mbuf. * Adjust uio_resid back * (it was adjusted * down by len bytes, * which we didn't end * up "copying" over). */ uio->uio_resid += len; break; } } else { SOCKBUF_UNLOCK(&so->so_rcv); *mp = m_copym(m, 0, len, M_WAITOK); SOCKBUF_LOCK(&so->so_rcv); } } sbcut_locked(&so->so_rcv, len); } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (so->so_oobmark) { if ((flags & MSG_PEEK) == 0) { so->so_oobmark -= len; if (so->so_oobmark == 0) { so->so_rcv.sb_state |= SBS_RCVATMARK; break; } } else { offset += len; if (offset == so->so_oobmark) break; } } if (flags & MSG_EOR) break; /* * If the MSG_WAITALL flag is set (for non-atomic socket), we * must not quit until "uio->uio_resid == 0" or an error * termination. If a signal/timeout occurs, return with a * short count but without error. Keep sockbuf locked * against other readers. */ while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && !sosendallatonce(so) && nextrecord == NULL) { SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE) break; /* * Notify the protocol that some data has been * drained before blocking. */ if (pr->pr_flags & PR_WANTRCVD) { SOCKBUF_UNLOCK(&so->so_rcv); VNET_SO_ASSERT(so); (*pr->pr_usrreqs->pru_rcvd)(so, flags); SOCKBUF_LOCK(&so->so_rcv); } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); /* * We could receive some data while was notifying * the protocol. Skip blocking in this case. */ if (so->so_rcv.sb_mb == NULL) { error = sbwait(&so->so_rcv); if (error) { SOCKBUF_UNLOCK(&so->so_rcv); goto release; } } m = so->so_rcv.sb_mb; if (m != NULL) nextrecord = m->m_nextpkt; } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (m != NULL && pr->pr_flags & PR_ATOMIC) { flags |= MSG_TRUNC; if ((flags & MSG_PEEK) == 0) (void) sbdroprecord_locked(&so->so_rcv); } if ((flags & MSG_PEEK) == 0) { if (m == NULL) { /* * First part is an inline SB_EMPTY_FIXUP(). Second * part makes sure sb_lastrecord is up-to-date if * there is still data in the socket buffer. */ so->so_rcv.sb_mb = nextrecord; if (so->so_rcv.sb_mb == NULL) { so->so_rcv.sb_mbtail = NULL; so->so_rcv.sb_lastrecord = NULL; } else if (nextrecord->m_nextpkt == NULL) so->so_rcv.sb_lastrecord = nextrecord; } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); /* * If soreceive() is being done from the socket callback, * then don't need to generate ACK to peer to update window, * since ACK will be generated on return to TCP. */ if (!(flags & MSG_SOCALLBCK) && (pr->pr_flags & PR_WANTRCVD)) { SOCKBUF_UNLOCK(&so->so_rcv); VNET_SO_ASSERT(so); (*pr->pr_usrreqs->pru_rcvd)(so, flags); SOCKBUF_LOCK(&so->so_rcv); } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (orig_resid == uio->uio_resid && orig_resid && (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { SOCKBUF_UNLOCK(&so->so_rcv); goto restart; } SOCKBUF_UNLOCK(&so->so_rcv); if (flagsp != NULL) *flagsp |= flags; release: sbunlock(&so->so_rcv); return (error); } /* * Optimized version of soreceive() for stream (TCP) sockets. */ int soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { int len = 0, error = 0, flags, oresid; struct sockbuf *sb; struct mbuf *m, *n = NULL; /* We only do stream sockets. */ if (so->so_type != SOCK_STREAM) return (EINVAL); if (psa != NULL) *psa = NULL; if (flagsp != NULL) flags = *flagsp &~ MSG_EOR; else flags = 0; if (controlp != NULL) *controlp = NULL; if (flags & MSG_OOB) return (soreceive_rcvoob(so, uio, flags)); if (mp0 != NULL) *mp0 = NULL; sb = &so->so_rcv; #ifdef KERN_TLS /* * KTLS store TLS records as records with a control message to * describe the framing. * * We check once here before acquiring locks to optimize the * common case. */ if (sb->sb_tls_info != NULL) return (soreceive_generic(so, psa, uio, mp0, controlp, flagsp)); #endif /* Prevent other readers from entering the socket. */ error = sblock(sb, SBLOCKWAIT(flags)); if (error) return (error); SOCKBUF_LOCK(sb); #ifdef KERN_TLS if (sb->sb_tls_info != NULL) { SOCKBUF_UNLOCK(sb); sbunlock(sb); return (soreceive_generic(so, psa, uio, mp0, controlp, flagsp)); } #endif /* Easy one, no space to copyout anything. */ if (uio->uio_resid == 0) { error = EINVAL; goto out; } oresid = uio->uio_resid; /* We will never ever get anything unless we are or were connected. */ if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { error = ENOTCONN; goto out; } restart: SOCKBUF_LOCK_ASSERT(&so->so_rcv); /* Abort if socket has reported problems. */ if (so->so_error) { if (sbavail(sb) > 0) goto deliver; if (oresid > uio->uio_resid) goto out; error = so->so_error; if (!(flags & MSG_PEEK)) so->so_error = 0; goto out; } /* Door is closed. Deliver what is left, if any. */ if (sb->sb_state & SBS_CANTRCVMORE) { if (sbavail(sb) > 0) goto deliver; else goto out; } /* Socket buffer is empty and we shall not block. */ if (sbavail(sb) == 0 && ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { error = EAGAIN; goto out; } /* Socket buffer got some data that we shall deliver now. */ if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)) || sbavail(sb) >= sb->sb_lowat || sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat) ) { goto deliver; } /* On MSG_WAITALL we must wait until all data or error arrives. */ if ((flags & MSG_WAITALL) && (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) goto deliver; /* * Wait and block until (more) data comes in. * NB: Drops the sockbuf lock during wait. */ error = sbwait(sb); if (error) goto out; goto restart; deliver: SOCKBUF_LOCK_ASSERT(&so->so_rcv); KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); /* Statistics. */ if (uio->uio_td) uio->uio_td->td_ru.ru_msgrcv++; /* Fill uio until full or current end of socket buffer is reached. */ len = min(uio->uio_resid, sbavail(sb)); if (mp0 != NULL) { /* Dequeue as many mbufs as possible. */ if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { if (*mp0 == NULL) *mp0 = sb->sb_mb; else m_cat(*mp0, sb->sb_mb); for (m = sb->sb_mb; m != NULL && m->m_len <= len; m = m->m_next) { KASSERT(!(m->m_flags & M_NOTAVAIL), ("%s: m %p not available", __func__, m)); len -= m->m_len; uio->uio_resid -= m->m_len; sbfree(sb, m); n = m; } n->m_next = NULL; sb->sb_mb = m; sb->sb_lastrecord = sb->sb_mb; if (sb->sb_mb == NULL) SB_EMPTY_FIXUP(sb); } /* Copy the remainder. */ if (len > 0) { KASSERT(sb->sb_mb != NULL, ("%s: len > 0 && sb->sb_mb empty", __func__)); m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); if (m == NULL) len = 0; /* Don't flush data from sockbuf. */ else uio->uio_resid -= len; if (*mp0 != NULL) m_cat(*mp0, m); else *mp0 = m; if (*mp0 == NULL) { error = ENOBUFS; goto out; } } } else { /* NB: Must unlock socket buffer as uiomove may sleep. */ SOCKBUF_UNLOCK(sb); error = m_mbuftouio(uio, sb->sb_mb, len); SOCKBUF_LOCK(sb); if (error) goto out; } SBLASTRECORDCHK(sb); SBLASTMBUFCHK(sb); /* * Remove the delivered data from the socket buffer unless we * were only peeking. */ if (!(flags & MSG_PEEK)) { if (len > 0) sbdrop_locked(sb, len); /* Notify protocol that we drained some data. */ if ((so->so_proto->pr_flags & PR_WANTRCVD) && (((flags & MSG_WAITALL) && uio->uio_resid > 0) || !(flags & MSG_SOCALLBCK))) { SOCKBUF_UNLOCK(sb); VNET_SO_ASSERT(so); (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); SOCKBUF_LOCK(sb); } } /* * For MSG_WAITALL we may have to loop again and wait for * more data to come in. */ if ((flags & MSG_WAITALL) && uio->uio_resid > 0) goto restart; out: SOCKBUF_LOCK_ASSERT(sb); SBLASTRECORDCHK(sb); SBLASTMBUFCHK(sb); SOCKBUF_UNLOCK(sb); sbunlock(sb); return (error); } /* * Optimized version of soreceive() for simple datagram cases from userspace. * Unlike in the stream case, we're able to drop a datagram if copyout() * fails, and because we handle datagrams atomically, we don't need to use a * sleep lock to prevent I/O interlacing. */ int soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { struct mbuf *m, *m2; int flags, error; ssize_t len; struct protosw *pr = so->so_proto; struct mbuf *nextrecord; if (psa != NULL) *psa = NULL; if (controlp != NULL) *controlp = NULL; if (flagsp != NULL) flags = *flagsp &~ MSG_EOR; else flags = 0; /* * For any complicated cases, fall back to the full * soreceive_generic(). */ if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) return (soreceive_generic(so, psa, uio, mp0, controlp, flagsp)); /* * Enforce restrictions on use. */ KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, ("soreceive_dgram: wantrcvd")); KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, ("soreceive_dgram: SBS_RCVATMARK")); KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, ("soreceive_dgram: P_CONNREQUIRED")); /* * Loop blocking while waiting for a datagram. */ SOCKBUF_LOCK(&so->so_rcv); while ((m = so->so_rcv.sb_mb) == NULL) { KASSERT(sbavail(&so->so_rcv) == 0, ("soreceive_dgram: sb_mb NULL but sbavail %u", sbavail(&so->so_rcv))); if (so->so_error) { error = so->so_error; so->so_error = 0; SOCKBUF_UNLOCK(&so->so_rcv); return (error); } if (so->so_rcv.sb_state & SBS_CANTRCVMORE || uio->uio_resid == 0) { SOCKBUF_UNLOCK(&so->so_rcv); return (0); } if ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO))) { SOCKBUF_UNLOCK(&so->so_rcv); return (EWOULDBLOCK); } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); error = sbwait(&so->so_rcv); if (error) { SOCKBUF_UNLOCK(&so->so_rcv); return (error); } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (uio->uio_td) uio->uio_td->td_ru.ru_msgrcv++; SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); nextrecord = m->m_nextpkt; if (nextrecord == NULL) { KASSERT(so->so_rcv.sb_lastrecord == m, ("soreceive_dgram: lastrecord != m")); } KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, ("soreceive_dgram: m_nextpkt != nextrecord")); /* * Pull 'm' and its chain off the front of the packet queue. */ so->so_rcv.sb_mb = NULL; sockbuf_pushsync(&so->so_rcv, nextrecord); /* * Walk 'm's chain and free that many bytes from the socket buffer. */ for (m2 = m; m2 != NULL; m2 = m2->m_next) sbfree(&so->so_rcv, m2); /* * Do a few last checks before we let go of the lock. */ SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); if (pr->pr_flags & PR_ADDR) { KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); if (psa != NULL) *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_NOWAIT); m = m_free(m); } if (m == NULL) { /* XXXRW: Can this happen? */ return (0); } /* * Packet to copyout() is now in 'm' and it is disconnected from the * queue. * * Process one or more MT_CONTROL mbufs present before any data mbufs * in the first mbuf chain on the socket buffer. We call into the * protocol to perform externalization (or freeing if controlp == * NULL). In some cases there can be only MT_CONTROL mbufs without * MT_DATA mbufs. */ if (m->m_type == MT_CONTROL) { struct mbuf *cm = NULL, *cmn; struct mbuf **cme = &cm; do { m2 = m->m_next; m->m_next = NULL; *cme = m; cme = &(*cme)->m_next; m = m2; } while (m != NULL && m->m_type == MT_CONTROL); while (cm != NULL) { cmn = cm->m_next; cm->m_next = NULL; if (pr->pr_domain->dom_externalize != NULL) { error = (*pr->pr_domain->dom_externalize) (cm, controlp, flags); } else if (controlp != NULL) *controlp = cm; else m_freem(cm); if (controlp != NULL) { while (*controlp != NULL) controlp = &(*controlp)->m_next; } cm = cmn; } } KASSERT(m == NULL || m->m_type == MT_DATA, ("soreceive_dgram: !data")); while (m != NULL && uio->uio_resid > 0) { len = uio->uio_resid; if (len > m->m_len) len = m->m_len; error = uiomove(mtod(m, char *), (int)len, uio); if (error) { m_freem(m); return (error); } if (len == m->m_len) m = m_free(m); else { m->m_data += len; m->m_len -= len; } } if (m != NULL) { flags |= MSG_TRUNC; m_freem(m); } if (flagsp != NULL) *flagsp |= flags; return (0); } int soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { int error; CURVNET_SET(so->so_vnet); if (!SOLISTENING(so)) error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0, controlp, flagsp)); else error = ENOTCONN; CURVNET_RESTORE(); return (error); } int soshutdown(struct socket *so, int how) { struct protosw *pr = so->so_proto; int error, soerror_enotconn; if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) return (EINVAL); soerror_enotconn = 0; if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { /* * POSIX mandates us to return ENOTCONN when shutdown(2) is * invoked on a datagram sockets, however historically we would * actually tear socket down. This is known to be leveraged by * some applications to unblock process waiting in recvXXX(2) * by other process that it shares that socket with. Try to meet * both backward-compatibility and POSIX requirements by forcing * ENOTCONN but still asking protocol to perform pru_shutdown(). */ if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) return (ENOTCONN); soerror_enotconn = 1; } if (SOLISTENING(so)) { if (how != SHUT_WR) { SOLISTEN_LOCK(so); so->so_error = ECONNABORTED; solisten_wakeup(so); /* unlocks so */ } goto done; } CURVNET_SET(so->so_vnet); if (pr->pr_usrreqs->pru_flush != NULL) (*pr->pr_usrreqs->pru_flush)(so, how); if (how != SHUT_WR) sorflush(so); if (how != SHUT_RD) { error = (*pr->pr_usrreqs->pru_shutdown)(so); wakeup(&so->so_timeo); CURVNET_RESTORE(); return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); } wakeup(&so->so_timeo); CURVNET_RESTORE(); done: return (soerror_enotconn ? ENOTCONN : 0); } void sorflush(struct socket *so) { struct sockbuf *sb = &so->so_rcv; struct protosw *pr = so->so_proto; struct socket aso; VNET_SO_ASSERT(so); /* * In order to avoid calling dom_dispose with the socket buffer mutex * held, and in order to generally avoid holding the lock for a long * time, we make a copy of the socket buffer and clear the original * (except locks, state). The new socket buffer copy won't have * initialized locks so we can only call routines that won't use or * assert those locks. * * Dislodge threads currently blocked in receive and wait to acquire * a lock against other simultaneous readers before clearing the * socket buffer. Don't let our acquire be interrupted by a signal * despite any existing socket disposition on interruptable waiting. */ socantrcvmore(so); (void) sblock(sb, SBL_WAIT | SBL_NOINTR); /* * Invalidate/clear most of the sockbuf structure, but leave selinfo * and mutex data unchanged. */ SOCKBUF_LOCK(sb); bzero(&aso, sizeof(aso)); aso.so_pcb = so->so_pcb; bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); bzero(&sb->sb_startzero, sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); SOCKBUF_UNLOCK(sb); sbunlock(sb); /* * Dispose of special rights and flush the copied socket. Don't call * any unsafe routines (that rely on locks being initialized) on aso. */ if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) (*pr->pr_domain->dom_dispose)(&aso); sbrelease_internal(&aso.so_rcv, so); } /* * Wrapper for Socket established helper hook. * Parameters: socket, context of the hook point, hook id. */ static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) { struct socket_hhook_data hhook_data = { .so = so, .hctx = hctx, .m = NULL, .status = 0 }; CURVNET_SET(so->so_vnet); HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); CURVNET_RESTORE(); /* Ugly but needed, since hhooks return void for now */ return (hhook_data.status); } /* * Perhaps this routine, and sooptcopyout(), below, ought to come in an * additional variant to handle the case where the option value needs to be * some kind of integer, but not a specific size. In addition to their use * here, these functions are also called by the protocol-level pr_ctloutput() * routines. */ int sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) { size_t valsize; /* * If the user gives us more than we wanted, we ignore it, but if we * don't get the minimum length the caller wants, we return EINVAL. * On success, sopt->sopt_valsize is set to however much we actually * retrieved. */ if ((valsize = sopt->sopt_valsize) < minlen) return EINVAL; if (valsize > len) sopt->sopt_valsize = valsize = len; if (sopt->sopt_td != NULL) return (copyin(sopt->sopt_val, buf, valsize)); bcopy(sopt->sopt_val, buf, valsize); return (0); } /* * Kernel version of setsockopt(2). * * XXX: optlen is size_t, not socklen_t */ int so_setsockopt(struct socket *so, int level, int optname, void *optval, size_t optlen) { struct sockopt sopt; sopt.sopt_level = level; sopt.sopt_name = optname; sopt.sopt_dir = SOPT_SET; sopt.sopt_val = optval; sopt.sopt_valsize = optlen; sopt.sopt_td = NULL; return (sosetopt(so, &sopt)); } int sosetopt(struct socket *so, struct sockopt *sopt) { int error, optval; struct linger l; struct timeval tv; sbintime_t val; uint32_t val32; #ifdef MAC struct mac extmac; #endif CURVNET_SET(so->so_vnet); error = 0; if (sopt->sopt_level != SOL_SOCKET) { if (so->so_proto->pr_ctloutput != NULL) error = (*so->so_proto->pr_ctloutput)(so, sopt); else error = ENOPROTOOPT; } else { switch (sopt->sopt_name) { case SO_ACCEPTFILTER: error = accept_filt_setopt(so, sopt); if (error) goto bad; break; case SO_LINGER: error = sooptcopyin(sopt, &l, sizeof l, sizeof l); if (error) goto bad; if (l.l_linger < 0 || l.l_linger > USHRT_MAX || l.l_linger > (INT_MAX / hz)) { error = EDOM; goto bad; } SOCK_LOCK(so); so->so_linger = l.l_linger; if (l.l_onoff) so->so_options |= SO_LINGER; else so->so_options &= ~SO_LINGER; SOCK_UNLOCK(so); break; case SO_DEBUG: case SO_KEEPALIVE: case SO_DONTROUTE: case SO_USELOOPBACK: case SO_BROADCAST: case SO_REUSEADDR: case SO_REUSEPORT: case SO_REUSEPORT_LB: case SO_OOBINLINE: case SO_TIMESTAMP: case SO_BINTIME: case SO_NOSIGPIPE: case SO_NO_DDP: case SO_NO_OFFLOAD: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) goto bad; SOCK_LOCK(so); if (optval) so->so_options |= sopt->sopt_name; else so->so_options &= ~sopt->sopt_name; SOCK_UNLOCK(so); break; case SO_SETFIB: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) goto bad; if (optval < 0 || optval >= rt_numfibs) { error = EINVAL; goto bad; } if (((so->so_proto->pr_domain->dom_family == PF_INET) || (so->so_proto->pr_domain->dom_family == PF_INET6) || (so->so_proto->pr_domain->dom_family == PF_ROUTE))) so->so_fibnum = optval; else so->so_fibnum = 0; break; case SO_USER_COOKIE: error = sooptcopyin(sopt, &val32, sizeof val32, sizeof val32); if (error) goto bad; so->so_user_cookie = val32; break; case SO_SNDBUF: case SO_RCVBUF: case SO_SNDLOWAT: case SO_RCVLOWAT: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) goto bad; /* * Values < 1 make no sense for any of these options, * so disallow them. */ if (optval < 1) { error = EINVAL; goto bad; } error = sbsetopt(so, sopt->sopt_name, optval); break; case SO_SNDTIMEO: case SO_RCVTIMEO: #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) { struct timeval32 tv32; error = sooptcopyin(sopt, &tv32, sizeof tv32, sizeof tv32); CP(tv32, tv, tv_sec); CP(tv32, tv, tv_usec); } else #endif error = sooptcopyin(sopt, &tv, sizeof tv, sizeof tv); if (error) goto bad; if (tv.tv_sec < 0 || tv.tv_usec < 0 || tv.tv_usec >= 1000000) { error = EDOM; goto bad; } if (tv.tv_sec > INT32_MAX) val = SBT_MAX; else val = tvtosbt(tv); switch (sopt->sopt_name) { case SO_SNDTIMEO: so->so_snd.sb_timeo = val; break; case SO_RCVTIMEO: so->so_rcv.sb_timeo = val; break; } break; case SO_LABEL: #ifdef MAC error = sooptcopyin(sopt, &extmac, sizeof extmac, sizeof extmac); if (error) goto bad; error = mac_setsockopt_label(sopt->sopt_td->td_ucred, so, &extmac); #else error = EOPNOTSUPP; #endif break; case SO_TS_CLOCK: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) goto bad; if (optval < 0 || optval > SO_TS_CLOCK_MAX) { error = EINVAL; goto bad; } so->so_ts_clock = optval; break; case SO_MAX_PACING_RATE: error = sooptcopyin(sopt, &val32, sizeof(val32), sizeof(val32)); if (error) goto bad; so->so_max_pacing_rate = val32; break; default: if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) error = hhook_run_socket(so, sopt, HHOOK_SOCKET_OPT); else error = ENOPROTOOPT; break; } if (error == 0 && so->so_proto->pr_ctloutput != NULL) (void)(*so->so_proto->pr_ctloutput)(so, sopt); } bad: CURVNET_RESTORE(); return (error); } /* * Helper routine for getsockopt. */ int sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) { int error; size_t valsize; error = 0; /* * Documented get behavior is that we always return a value, possibly * truncated to fit in the user's buffer. Traditional behavior is * that we always tell the user precisely how much we copied, rather * than something useful like the total amount we had available for * her. Note that this interface is not idempotent; the entire * answer must be generated ahead of time. */ valsize = min(len, sopt->sopt_valsize); sopt->sopt_valsize = valsize; if (sopt->sopt_val != NULL) { if (sopt->sopt_td != NULL) error = copyout(buf, sopt->sopt_val, valsize); else bcopy(buf, sopt->sopt_val, valsize); } return (error); } int sogetopt(struct socket *so, struct sockopt *sopt) { int error, optval; struct linger l; struct timeval tv; #ifdef MAC struct mac extmac; #endif CURVNET_SET(so->so_vnet); error = 0; if (sopt->sopt_level != SOL_SOCKET) { if (so->so_proto->pr_ctloutput != NULL) error = (*so->so_proto->pr_ctloutput)(so, sopt); else error = ENOPROTOOPT; CURVNET_RESTORE(); return (error); } else { switch (sopt->sopt_name) { case SO_ACCEPTFILTER: error = accept_filt_getopt(so, sopt); break; case SO_LINGER: SOCK_LOCK(so); l.l_onoff = so->so_options & SO_LINGER; l.l_linger = so->so_linger; SOCK_UNLOCK(so); error = sooptcopyout(sopt, &l, sizeof l); break; case SO_USELOOPBACK: case SO_DONTROUTE: case SO_DEBUG: case SO_KEEPALIVE: case SO_REUSEADDR: case SO_REUSEPORT: case SO_REUSEPORT_LB: case SO_BROADCAST: case SO_OOBINLINE: case SO_ACCEPTCONN: case SO_TIMESTAMP: case SO_BINTIME: case SO_NOSIGPIPE: case SO_NO_DDP: case SO_NO_OFFLOAD: optval = so->so_options & sopt->sopt_name; integer: error = sooptcopyout(sopt, &optval, sizeof optval); break; case SO_DOMAIN: optval = so->so_proto->pr_domain->dom_family; goto integer; case SO_TYPE: optval = so->so_type; goto integer; case SO_PROTOCOL: optval = so->so_proto->pr_protocol; goto integer; case SO_ERROR: SOCK_LOCK(so); optval = so->so_error; so->so_error = 0; SOCK_UNLOCK(so); goto integer; case SO_SNDBUF: optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : so->so_snd.sb_hiwat; goto integer; case SO_RCVBUF: optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : so->so_rcv.sb_hiwat; goto integer; case SO_SNDLOWAT: optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : so->so_snd.sb_lowat; goto integer; case SO_RCVLOWAT: optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : so->so_rcv.sb_lowat; goto integer; case SO_SNDTIMEO: case SO_RCVTIMEO: tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? so->so_snd.sb_timeo : so->so_rcv.sb_timeo); #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) { struct timeval32 tv32; CP(tv, tv32, tv_sec); CP(tv, tv32, tv_usec); error = sooptcopyout(sopt, &tv32, sizeof tv32); } else #endif error = sooptcopyout(sopt, &tv, sizeof tv); break; case SO_LABEL: #ifdef MAC error = sooptcopyin(sopt, &extmac, sizeof(extmac), sizeof(extmac)); if (error) goto bad; error = mac_getsockopt_label(sopt->sopt_td->td_ucred, so, &extmac); if (error) goto bad; error = sooptcopyout(sopt, &extmac, sizeof extmac); #else error = EOPNOTSUPP; #endif break; case SO_PEERLABEL: #ifdef MAC error = sooptcopyin(sopt, &extmac, sizeof(extmac), sizeof(extmac)); if (error) goto bad; error = mac_getsockopt_peerlabel( sopt->sopt_td->td_ucred, so, &extmac); if (error) goto bad; error = sooptcopyout(sopt, &extmac, sizeof extmac); #else error = EOPNOTSUPP; #endif break; case SO_LISTENQLIMIT: optval = SOLISTENING(so) ? so->sol_qlimit : 0; goto integer; case SO_LISTENQLEN: optval = SOLISTENING(so) ? so->sol_qlen : 0; goto integer; case SO_LISTENINCQLEN: optval = SOLISTENING(so) ? so->sol_incqlen : 0; goto integer; case SO_TS_CLOCK: optval = so->so_ts_clock; goto integer; case SO_MAX_PACING_RATE: optval = so->so_max_pacing_rate; goto integer; default: if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) error = hhook_run_socket(so, sopt, HHOOK_SOCKET_OPT); else error = ENOPROTOOPT; break; } } #ifdef MAC bad: #endif CURVNET_RESTORE(); return (error); } int soopt_getm(struct sockopt *sopt, struct mbuf **mp) { struct mbuf *m, *m_prev; int sopt_size = sopt->sopt_valsize; MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); if (m == NULL) return ENOBUFS; if (sopt_size > MLEN) { MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); if ((m->m_flags & M_EXT) == 0) { m_free(m); return ENOBUFS; } m->m_len = min(MCLBYTES, sopt_size); } else { m->m_len = min(MLEN, sopt_size); } sopt_size -= m->m_len; *mp = m; m_prev = m; while (sopt_size) { MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); if (m == NULL) { m_freem(*mp); return ENOBUFS; } if (sopt_size > MLEN) { MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : M_NOWAIT); if ((m->m_flags & M_EXT) == 0) { m_freem(m); m_freem(*mp); return ENOBUFS; } m->m_len = min(MCLBYTES, sopt_size); } else { m->m_len = min(MLEN, sopt_size); } sopt_size -= m->m_len; m_prev->m_next = m; m_prev = m; } return (0); } int soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) { struct mbuf *m0 = m; if (sopt->sopt_val == NULL) return (0); while (m != NULL && sopt->sopt_valsize >= m->m_len) { if (sopt->sopt_td != NULL) { int error; error = copyin(sopt->sopt_val, mtod(m, char *), m->m_len); if (error != 0) { m_freem(m0); return(error); } } else bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); sopt->sopt_valsize -= m->m_len; sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; m = m->m_next; } if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ panic("ip6_sooptmcopyin"); return (0); } int soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) { struct mbuf *m0 = m; size_t valsize = 0; if (sopt->sopt_val == NULL) return (0); while (m != NULL && sopt->sopt_valsize >= m->m_len) { if (sopt->sopt_td != NULL) { int error; error = copyout(mtod(m, char *), sopt->sopt_val, m->m_len); if (error != 0) { m_freem(m0); return(error); } } else bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); sopt->sopt_valsize -= m->m_len; sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; valsize += m->m_len; m = m->m_next; } if (m != NULL) { /* enough soopt buffer should be given from user-land */ m_freem(m0); return(EINVAL); } sopt->sopt_valsize = valsize; return (0); } /* * sohasoutofband(): protocol notifies socket layer of the arrival of new * out-of-band data, which will then notify socket consumers. */ void sohasoutofband(struct socket *so) { if (so->so_sigio != NULL) pgsigio(&so->so_sigio, SIGURG, 0); selwakeuppri(&so->so_rdsel, PSOCK); } int sopoll(struct socket *so, int events, struct ucred *active_cred, struct thread *td) { /* * We do not need to set or assert curvnet as long as everyone uses * sopoll_generic(). */ return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, td)); } int sopoll_generic(struct socket *so, int events, struct ucred *active_cred, struct thread *td) { int revents; SOCK_LOCK(so); if (SOLISTENING(so)) { if (!(events & (POLLIN | POLLRDNORM))) revents = 0; else if (!TAILQ_EMPTY(&so->sol_comp)) revents = events & (POLLIN | POLLRDNORM); else if ((events & POLLINIGNEOF) == 0 && so->so_error) revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; else { selrecord(td, &so->so_rdsel); revents = 0; } } else { revents = 0; SOCKBUF_LOCK(&so->so_snd); SOCKBUF_LOCK(&so->so_rcv); if (events & (POLLIN | POLLRDNORM)) if (soreadabledata(so)) revents |= events & (POLLIN | POLLRDNORM); if (events & (POLLOUT | POLLWRNORM)) if (sowriteable(so)) revents |= events & (POLLOUT | POLLWRNORM); if (events & (POLLPRI | POLLRDBAND)) if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK)) revents |= events & (POLLPRI | POLLRDBAND); if ((events & POLLINIGNEOF) == 0) { if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { revents |= events & (POLLIN | POLLRDNORM); if (so->so_snd.sb_state & SBS_CANTSENDMORE) revents |= POLLHUP; } } if (revents == 0) { if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { selrecord(td, &so->so_rdsel); so->so_rcv.sb_flags |= SB_SEL; } if (events & (POLLOUT | POLLWRNORM)) { selrecord(td, &so->so_wrsel); so->so_snd.sb_flags |= SB_SEL; } } SOCKBUF_UNLOCK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_snd); } SOCK_UNLOCK(so); return (revents); } int soo_kqfilter(struct file *fp, struct knote *kn) { struct socket *so = kn->kn_fp->f_data; struct sockbuf *sb; struct knlist *knl; switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &soread_filtops; knl = &so->so_rdsel.si_note; sb = &so->so_rcv; break; case EVFILT_WRITE: kn->kn_fop = &sowrite_filtops; knl = &so->so_wrsel.si_note; sb = &so->so_snd; break; case EVFILT_EMPTY: kn->kn_fop = &soempty_filtops; knl = &so->so_wrsel.si_note; sb = &so->so_snd; break; default: return (EINVAL); } SOCK_LOCK(so); if (SOLISTENING(so)) { knlist_add(knl, kn, 1); } else { SOCKBUF_LOCK(sb); knlist_add(knl, kn, 1); sb->sb_flags |= SB_KNOTE; SOCKBUF_UNLOCK(sb); } SOCK_UNLOCK(so); return (0); } /* * Some routines that return EOPNOTSUPP for entry points that are not * supported by a protocol. Fill in as needed. */ int pru_accept_notsupp(struct socket *so, struct sockaddr **nam) { return EOPNOTSUPP; } int pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) { return EOPNOTSUPP; } int pru_attach_notsupp(struct socket *so, int proto, struct thread *td) { return EOPNOTSUPP; } int pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) { return EOPNOTSUPP; } int pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { return EOPNOTSUPP; } int pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) { return EOPNOTSUPP; } int pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { return EOPNOTSUPP; } int pru_connect2_notsupp(struct socket *so1, struct socket *so2) { return EOPNOTSUPP; } int pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, struct ifnet *ifp, struct thread *td) { return EOPNOTSUPP; } int pru_disconnect_notsupp(struct socket *so) { return EOPNOTSUPP; } int pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) { return EOPNOTSUPP; } int pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) { return EOPNOTSUPP; } int pru_rcvd_notsupp(struct socket *so, int flags) { return EOPNOTSUPP; } int pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) { return EOPNOTSUPP; } int pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *td) { return EOPNOTSUPP; } int pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) { return (EOPNOTSUPP); } /* * This isn't really a ``null'' operation, but it's the default one and * doesn't do anything destructive. */ int pru_sense_null(struct socket *so, struct stat *sb) { sb->st_blksize = so->so_snd.sb_hiwat; return 0; } int pru_shutdown_notsupp(struct socket *so) { return EOPNOTSUPP; } int pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) { return EOPNOTSUPP; } int pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { return EOPNOTSUPP; } int pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { return EOPNOTSUPP; } int pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, struct thread *td) { return EOPNOTSUPP; } static void filt_sordetach(struct knote *kn) { struct socket *so = kn->kn_fp->f_data; so_rdknl_lock(so); knlist_remove(&so->so_rdsel.si_note, kn, 1); if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) so->so_rcv.sb_flags &= ~SB_KNOTE; so_rdknl_unlock(so); } /*ARGSUSED*/ static int filt_soread(struct knote *kn, long hint) { struct socket *so; so = kn->kn_fp->f_data; if (SOLISTENING(so)) { SOCK_LOCK_ASSERT(so); kn->kn_data = so->sol_qlen; if (so->so_error) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; return (1); } return (!TAILQ_EMPTY(&so->sol_comp)); } SOCKBUF_LOCK_ASSERT(&so->so_rcv); kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; return (1); } else if (so->so_error) /* temporary udp error */ return (1); if (kn->kn_sfflags & NOTE_LOWAT) { if (kn->kn_data >= kn->kn_sdata) return (1); } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) return (1); /* This hook returning non-zero indicates an event, not error */ return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); } static void filt_sowdetach(struct knote *kn) { struct socket *so = kn->kn_fp->f_data; so_wrknl_lock(so); knlist_remove(&so->so_wrsel.si_note, kn, 1); if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) so->so_snd.sb_flags &= ~SB_KNOTE; so_wrknl_unlock(so); } /*ARGSUSED*/ static int filt_sowrite(struct knote *kn, long hint) { struct socket *so; so = kn->kn_fp->f_data; if (SOLISTENING(so)) return (0); SOCKBUF_LOCK_ASSERT(&so->so_snd); kn->kn_data = sbspace(&so->so_snd); hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); if (so->so_snd.sb_state & SBS_CANTSENDMORE) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; return (1); } else if (so->so_error) /* temporary udp error */ return (1); else if (((so->so_state & SS_ISCONNECTED) == 0) && (so->so_proto->pr_flags & PR_CONNREQUIRED)) return (0); else if (kn->kn_sfflags & NOTE_LOWAT) return (kn->kn_data >= kn->kn_sdata); else return (kn->kn_data >= so->so_snd.sb_lowat); } static int filt_soempty(struct knote *kn, long hint) { struct socket *so; so = kn->kn_fp->f_data; if (SOLISTENING(so)) return (1); SOCKBUF_LOCK_ASSERT(&so->so_snd); kn->kn_data = sbused(&so->so_snd); if (kn->kn_data == 0) return (1); else return (0); } int socheckuid(struct socket *so, uid_t uid) { if (so == NULL) return (EPERM); if (so->so_cred->cr_uid != uid) return (EPERM); return (0); } /* * These functions are used by protocols to notify the socket layer (and its * consumers) of state changes in the sockets driven by protocol-side events. */ /* * Procedures to manipulate state flags of socket and do appropriate wakeups. * * Normal sequence from the active (originating) side is that * soisconnecting() is called during processing of connect() call, resulting * in an eventual call to soisconnected() if/when the connection is * established. When the connection is torn down soisdisconnecting() is * called during processing of disconnect() call, and soisdisconnected() is * called when the connection to the peer is totally severed. The semantics * of these routines are such that connectionless protocols can call * soisconnected() and soisdisconnected() only, bypassing the in-progress * calls when setting up a ``connection'' takes no time. * * From the passive side, a socket is created with two queues of sockets: * so_incomp for connections in progress and so_comp for connections already * made and awaiting user acceptance. As a protocol is preparing incoming * connections, it creates a socket structure queued on so_incomp by calling * sonewconn(). When the connection is established, soisconnected() is * called, and transfers the socket structure to so_comp, making it available * to accept(). * * If a socket is closed with sockets on either so_incomp or so_comp, these * sockets are dropped. * * If higher-level protocols are implemented in the kernel, the wakeups done * here will sometimes cause software-interrupt process scheduling. */ void soisconnecting(struct socket *so) { SOCK_LOCK(so); so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); so->so_state |= SS_ISCONNECTING; SOCK_UNLOCK(so); } void soisconnected(struct socket *so) { SOCK_LOCK(so); so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); so->so_state |= SS_ISCONNECTED; if (so->so_qstate == SQ_INCOMP) { struct socket *head = so->so_listen; int ret; KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); /* * Promoting a socket from incomplete queue to complete, we * need to go through reverse order of locking. We first do * trylock, and if that doesn't succeed, we go the hard way * leaving a reference and rechecking consistency after proper * locking. */ if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { soref(head); SOCK_UNLOCK(so); SOLISTEN_LOCK(head); SOCK_LOCK(so); if (__predict_false(head != so->so_listen)) { /* * The socket went off the listen queue, * should be lost race to close(2) of sol. * The socket is about to soabort(). */ SOCK_UNLOCK(so); sorele(head); return; } /* Not the last one, as so holds a ref. */ refcount_release(&head->so_count); } again: if ((so->so_options & SO_ACCEPTFILTER) == 0) { TAILQ_REMOVE(&head->sol_incomp, so, so_list); head->sol_incqlen--; TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); head->sol_qlen++; so->so_qstate = SQ_COMP; SOCK_UNLOCK(so); solisten_wakeup(head); /* unlocks */ } else { SOCKBUF_LOCK(&so->so_rcv); soupcall_set(so, SO_RCV, head->sol_accept_filter->accf_callback, head->sol_accept_filter_arg); so->so_options &= ~SO_ACCEPTFILTER; ret = head->sol_accept_filter->accf_callback(so, head->sol_accept_filter_arg, M_NOWAIT); if (ret == SU_ISCONNECTED) { soupcall_clear(so, SO_RCV); SOCKBUF_UNLOCK(&so->so_rcv); goto again; } SOCKBUF_UNLOCK(&so->so_rcv); SOCK_UNLOCK(so); SOLISTEN_UNLOCK(head); } return; } SOCK_UNLOCK(so); wakeup(&so->so_timeo); sorwakeup(so); sowwakeup(so); } void soisdisconnecting(struct socket *so) { SOCK_LOCK(so); so->so_state &= ~SS_ISCONNECTING; so->so_state |= SS_ISDISCONNECTING; if (!SOLISTENING(so)) { SOCKBUF_LOCK(&so->so_rcv); socantrcvmore_locked(so); SOCKBUF_LOCK(&so->so_snd); socantsendmore_locked(so); } SOCK_UNLOCK(so); wakeup(&so->so_timeo); } void soisdisconnected(struct socket *so) { SOCK_LOCK(so); /* * There is at least one reader of so_state that does not * acquire socket lock, namely soreceive_generic(). Ensure * that it never sees all flags that track connection status * cleared, by ordering the update with a barrier semantic of * our release thread fence. */ so->so_state |= SS_ISDISCONNECTED; atomic_thread_fence_rel(); so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); if (!SOLISTENING(so)) { SOCK_UNLOCK(so); SOCKBUF_LOCK(&so->so_rcv); socantrcvmore_locked(so); SOCKBUF_LOCK(&so->so_snd); sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); socantsendmore_locked(so); } else SOCK_UNLOCK(so); wakeup(&so->so_timeo); } /* * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. */ struct sockaddr * sodupsockaddr(const struct sockaddr *sa, int mflags) { struct sockaddr *sa2; sa2 = malloc(sa->sa_len, M_SONAME, mflags); if (sa2) bcopy(sa, sa2, sa->sa_len); return sa2; } /* * Register per-socket destructor. */ void sodtor_set(struct socket *so, so_dtor_t *func) { SOCK_LOCK_ASSERT(so); so->so_dtor = func; } /* * Register per-socket buffer upcalls. */ void soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) { struct sockbuf *sb; KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); switch (which) { case SO_RCV: sb = &so->so_rcv; break; case SO_SND: sb = &so->so_snd; break; default: panic("soupcall_set: bad which"); } SOCKBUF_LOCK_ASSERT(sb); sb->sb_upcall = func; sb->sb_upcallarg = arg; sb->sb_flags |= SB_UPCALL; } void soupcall_clear(struct socket *so, int which) { struct sockbuf *sb; KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); switch (which) { case SO_RCV: sb = &so->so_rcv; break; case SO_SND: sb = &so->so_snd; break; default: panic("soupcall_clear: bad which"); } SOCKBUF_LOCK_ASSERT(sb); KASSERT(sb->sb_upcall != NULL, ("%s: so %p no upcall to clear", __func__, so)); sb->sb_upcall = NULL; sb->sb_upcallarg = NULL; sb->sb_flags &= ~SB_UPCALL; } void solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) { SOLISTEN_LOCK_ASSERT(so); so->sol_upcall = func; so->sol_upcallarg = arg; } static void so_rdknl_lock(void *arg) { struct socket *so = arg; if (SOLISTENING(so)) SOCK_LOCK(so); else SOCKBUF_LOCK(&so->so_rcv); } static void so_rdknl_unlock(void *arg) { struct socket *so = arg; if (SOLISTENING(so)) SOCK_UNLOCK(so); else SOCKBUF_UNLOCK(&so->so_rcv); } static void -so_rdknl_assert_locked(void *arg) +so_rdknl_assert_lock(void *arg, int what) { struct socket *so = arg; - if (SOLISTENING(so)) - SOCK_LOCK_ASSERT(so); - else - SOCKBUF_LOCK_ASSERT(&so->so_rcv); -} - -static void -so_rdknl_assert_unlocked(void *arg) -{ - struct socket *so = arg; - - if (SOLISTENING(so)) - SOCK_UNLOCK_ASSERT(so); - else - SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); + if (what == LA_LOCKED) { + if (SOLISTENING(so)) + SOCK_LOCK_ASSERT(so); + else + SOCKBUF_LOCK_ASSERT(&so->so_rcv); + } else { + if (SOLISTENING(so)) + SOCK_UNLOCK_ASSERT(so); + else + SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); + } } static void so_wrknl_lock(void *arg) { struct socket *so = arg; if (SOLISTENING(so)) SOCK_LOCK(so); else SOCKBUF_LOCK(&so->so_snd); } static void so_wrknl_unlock(void *arg) { struct socket *so = arg; if (SOLISTENING(so)) SOCK_UNLOCK(so); else SOCKBUF_UNLOCK(&so->so_snd); } static void -so_wrknl_assert_locked(void *arg) -{ - struct socket *so = arg; - - if (SOLISTENING(so)) - SOCK_LOCK_ASSERT(so); - else - SOCKBUF_LOCK_ASSERT(&so->so_snd); -} - -static void -so_wrknl_assert_unlocked(void *arg) +so_wrknl_assert_lock(void *arg, int what) { struct socket *so = arg; - if (SOLISTENING(so)) - SOCK_UNLOCK_ASSERT(so); - else - SOCKBUF_UNLOCK_ASSERT(&so->so_snd); + if (what == LA_LOCKED) { + if (SOLISTENING(so)) + SOCK_LOCK_ASSERT(so); + else + SOCKBUF_LOCK_ASSERT(&so->so_snd); + } else { + if (SOLISTENING(so)) + SOCK_UNLOCK_ASSERT(so); + else + SOCKBUF_UNLOCK_ASSERT(&so->so_snd); + } } /* * Create an external-format (``xsocket'') structure using the information in * the kernel-format socket structure pointed to by so. This is done to * reduce the spew of irrelevant information over this interface, to isolate * user code from changes in the kernel structure, and potentially to provide * information-hiding if we decide that some of this information should be * hidden from users. */ void sotoxsocket(struct socket *so, struct xsocket *xso) { bzero(xso, sizeof(*xso)); xso->xso_len = sizeof *xso; xso->xso_so = (uintptr_t)so; xso->so_type = so->so_type; xso->so_options = so->so_options; xso->so_linger = so->so_linger; xso->so_state = so->so_state; xso->so_pcb = (uintptr_t)so->so_pcb; xso->xso_protocol = so->so_proto->pr_protocol; xso->xso_family = so->so_proto->pr_domain->dom_family; xso->so_timeo = so->so_timeo; xso->so_error = so->so_error; xso->so_uid = so->so_cred->cr_uid; xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; if (SOLISTENING(so)) { xso->so_qlen = so->sol_qlen; xso->so_incqlen = so->sol_incqlen; xso->so_qlimit = so->sol_qlimit; xso->so_oobmark = 0; } else { xso->so_state |= so->so_qstate; xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; xso->so_oobmark = so->so_oobmark; sbtoxsockbuf(&so->so_snd, &xso->so_snd); sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); } } struct sockbuf * so_sockbuf_rcv(struct socket *so) { return (&so->so_rcv); } struct sockbuf * so_sockbuf_snd(struct socket *so) { return (&so->so_snd); } int so_state_get(const struct socket *so) { return (so->so_state); } void so_state_set(struct socket *so, int val) { so->so_state = val; } int so_options_get(const struct socket *so) { return (so->so_options); } void so_options_set(struct socket *so, int val) { so->so_options = val; } int so_error_get(const struct socket *so) { return (so->so_error); } void so_error_set(struct socket *so, int val) { so->so_error = val; } int so_linger_get(const struct socket *so) { return (so->so_linger); } void so_linger_set(struct socket *so, int val) { KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), ("%s: val %d out of range", __func__, val)); so->so_linger = val; } struct protosw * so_protosw_get(const struct socket *so) { return (so->so_proto); } void so_protosw_set(struct socket *so, struct protosw *val) { so->so_proto = val; } void so_sorwakeup(struct socket *so) { sorwakeup(so); } void so_sowwakeup(struct socket *so) { sowwakeup(so); } void so_sorwakeup_locked(struct socket *so) { sorwakeup_locked(so); } void so_sowwakeup_locked(struct socket *so) { sowwakeup_locked(so); } void so_lock(struct socket *so) { SOCK_LOCK(so); } void so_unlock(struct socket *so) { SOCK_UNLOCK(so); } diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c index 1a6536faa646..6045c704b227 100644 --- a/sys/kern/vfs_subr.c +++ b/sys/kern/vfs_subr.c @@ -1,6800 +1,6792 @@ /*- * 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 bool vhold_recycle_free(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 vfs_knl_assert_lock(void *arg, int what); 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; MALLOC_DEFINE(M_VNODEPOLL, "VN POLL", "vnode poll"); __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; /* * Check vhold_recycle_free for an explanation. */ vp->v_holdcnt = VHOLD_NO_SMR; vp->v_type = VNON; 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); /* * 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; object = atomic_load_ptr(&vp->v_object); if (object == NULL || object->resident_page_count > trigger) { goto next_iter; } /* * Handle races against vnode allocation. Filesystems lock the * vnode some time after it gets returned from getnewvnode, * despite type and hold count being manipulated earlier. * Resorting to checking v_mount restores guarantees present * before the global list was reworked to contain all vnodes. */ if (!VI_TRYLOCK(vp)) goto next_iter; if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { VI_UNLOCK(vp); goto next_iter; } if (vp->v_mount == NULL) { 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; vp = mvp; for (;;) { if (count == 0) { break; } 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; if (vp->v_holdcnt > 0) 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. */ if (mnt_op != NULL && (mp = vp->v_mount) != NULL && mp->mnt_op != mnt_op) { continue; } if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { continue; } if (!vhold_recycle_free(vp)) continue; TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); if (vtryrecycle(vp) == 0) count--; mtx_lock(&vnode_list_mtx); vp = mvp; } 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); vdrop(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); vdrop(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); vdropl(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); vdropl(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")); VNASSERT((vp->v_iflag & (VI_DOINGINACT | VI_OWEINACT)) == 0, vp, ("Leaked inactivation")); 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 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); } } } /* * Hold a free vnode for recycling. * * Note: vnode_init references this comment. * * Attempts to recycle only need the global vnode list lock and have no use for * SMR. * * However, vnodes get inserted into the global list before they get fully * initialized and stay there until UMA decides to free the memory. This in * particular means the target can be found before it becomes usable and after * it becomes recycled. Picking up such vnodes is guarded with v_holdcnt set to * VHOLD_NO_SMR. * * Note: the vnode may gain more references after we transition the count 0->1. */ static bool vhold_recycle_free(struct vnode *vp) { int count; mtx_assert(&vnode_list_mtx, MA_OWNED); 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 (count > 0) { return (false); } if (atomic_fcmpset_int(&vp->v_holdcnt, &count, count + 1)) { 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); 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, doinginact, 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(). * * It could be that VOP_INACTIVE() requested reclamation, in * which case we should avoid recursion, so check * VI_DOINGINACT. This is not precise but good enough. */ active = vp->v_usecount > 0; oweinact = (vp->v_iflag & VI_OWEINACT) != 0; doinginact = (vp->v_iflag & VI_DOINGINACT) != 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); } cache_purge_vgone(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) && !doinginact) { 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); /* * 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; int lkflags, objflags; bool seen_defer; 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); free(vi, M_VNODEPOLL); } 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 = malloc(sizeof(*vi), M_VNODEPOLL, 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); + vfs_knlunlock, vfs_knl_assert_lock); 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, 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. */ error = priv_check_cred_vfs_lookup_nomac(cred); switch (error) { case 0: return (0); case EAGAIN: /* * MAC modules present. */ return (EAGAIN); 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); } #ifdef DEBUG_VFS_LOCKS void vop_mkdir_debugpost(void *ap, int rc) { struct vop_mkdir_args *a; a = ap; if (!rc) cache_validate(a->a_dvp, *a->a_vpp, a->a_cnp); } #endif 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_read_pgcache_post(void *ap, int rc) { struct vop_read_pgcache_args *a = ap; if (!rc) VFS_KNOTE_UNLOCKED(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) +vfs_knl_assert_lock(void *arg, int what) { #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"); + if (what == LA_LOCKED) + ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); + else + 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); } diff --git a/sys/sys/event.h b/sys/sys/event.h index f7e822a726a5..b3106e6fa874 100644 --- a/sys/sys/event.h +++ b/sys/sys/event.h @@ -1,368 +1,366 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1999,2000,2001 Jonathan Lemon * 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. * * $FreeBSD$ */ #ifndef _SYS_EVENT_H_ #define _SYS_EVENT_H_ #include #include #define EVFILT_READ (-1) #define EVFILT_WRITE (-2) #define EVFILT_AIO (-3) /* attached to aio requests */ #define EVFILT_VNODE (-4) /* attached to vnodes */ #define EVFILT_PROC (-5) /* attached to struct proc */ #define EVFILT_SIGNAL (-6) /* attached to struct proc */ #define EVFILT_TIMER (-7) /* timers */ #define EVFILT_PROCDESC (-8) /* attached to process descriptors */ #define EVFILT_FS (-9) /* filesystem events */ #define EVFILT_LIO (-10) /* attached to lio requests */ #define EVFILT_USER (-11) /* User events */ #define EVFILT_SENDFILE (-12) /* attached to sendfile requests */ #define EVFILT_EMPTY (-13) /* empty send socket buf */ #define EVFILT_SYSCOUNT 13 #if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L #define EV_SET(kevp_, a, b, c, d, e, f) do { \ *(kevp_) = (struct kevent){ \ .ident = (a), \ .filter = (b), \ .flags = (c), \ .fflags = (d), \ .data = (e), \ .udata = (f), \ .ext = {0}, \ }; \ } while(0) #else /* Pre-C99 or not STDC (e.g., C++) */ /* The definition of the local variable kevp could possibly conflict * with a user-defined value passed in parameters a-f. */ #define EV_SET(kevp_, a, b, c, d, e, f) do { \ struct kevent *kevp = (kevp_); \ (kevp)->ident = (a); \ (kevp)->filter = (b); \ (kevp)->flags = (c); \ (kevp)->fflags = (d); \ (kevp)->data = (e); \ (kevp)->udata = (f); \ (kevp)->ext[0] = 0; \ (kevp)->ext[1] = 0; \ (kevp)->ext[2] = 0; \ (kevp)->ext[3] = 0; \ } while(0) #endif struct kevent { __uintptr_t ident; /* identifier for this event */ short filter; /* filter for event */ unsigned short flags; /* action flags for kqueue */ unsigned int fflags; /* filter flag value */ __int64_t data; /* filter data value */ void *udata; /* opaque user data identifier */ __uint64_t ext[4]; /* extensions */ }; #if defined(_WANT_FREEBSD11_KEVENT) /* Older structure used in FreeBSD 11.x and older. */ struct kevent_freebsd11 { __uintptr_t ident; /* identifier for this event */ short filter; /* filter for event */ unsigned short flags; unsigned int fflags; __intptr_t data; void *udata; /* opaque user data identifier */ }; #endif #if defined(_WANT_KEVENT32) || (defined(_KERNEL) && defined(__LP64__)) struct kevent32 { uint32_t ident; /* identifier for this event */ short filter; /* filter for event */ u_short flags; u_int fflags; #ifndef __amd64__ uint32_t pad0; #endif uint32_t data1, data2; uint32_t udata; /* opaque user data identifier */ #ifndef __amd64__ uint32_t pad1; #endif uint32_t ext64[8]; }; #ifdef _WANT_FREEBSD11_KEVENT struct kevent32_freebsd11 { u_int32_t ident; /* identifier for this event */ short filter; /* filter for event */ u_short flags; u_int fflags; int32_t data; u_int32_t udata; /* opaque user data identifier */ }; #endif #endif /* actions */ #define EV_ADD 0x0001 /* add event to kq (implies enable) */ #define EV_DELETE 0x0002 /* delete event from kq */ #define EV_ENABLE 0x0004 /* enable event */ #define EV_DISABLE 0x0008 /* disable event (not reported) */ #define EV_FORCEONESHOT 0x0100 /* enable _ONESHOT and force trigger */ /* flags */ #define EV_ONESHOT 0x0010 /* only report one occurrence */ #define EV_CLEAR 0x0020 /* clear event state after reporting */ #define EV_RECEIPT 0x0040 /* force EV_ERROR on success, data=0 */ #define EV_DISPATCH 0x0080 /* disable event after reporting */ #define EV_SYSFLAGS 0xF000 /* reserved by system */ #define EV_DROP 0x1000 /* note should be dropped */ #define EV_FLAG1 0x2000 /* filter-specific flag */ #define EV_FLAG2 0x4000 /* filter-specific flag */ /* returned values */ #define EV_EOF 0x8000 /* EOF detected */ #define EV_ERROR 0x4000 /* error, data contains errno */ /* * data/hint flags/masks for EVFILT_USER, shared with userspace * * On input, the top two bits of fflags specifies how the lower twenty four * bits should be applied to the stored value of fflags. * * On output, the top two bits will always be set to NOTE_FFNOP and the * remaining twenty four bits will contain the stored fflags value. */ #define NOTE_FFNOP 0x00000000 /* ignore input fflags */ #define NOTE_FFAND 0x40000000 /* AND fflags */ #define NOTE_FFOR 0x80000000 /* OR fflags */ #define NOTE_FFCOPY 0xc0000000 /* copy fflags */ #define NOTE_FFCTRLMASK 0xc0000000 /* masks for operations */ #define NOTE_FFLAGSMASK 0x00ffffff #define NOTE_TRIGGER 0x01000000 /* Cause the event to be triggered for output. */ /* * data/hint flags for EVFILT_{READ|WRITE}, shared with userspace */ #define NOTE_LOWAT 0x0001 /* low water mark */ #define NOTE_FILE_POLL 0x0002 /* behave like poll() */ /* * data/hint flags for EVFILT_VNODE, shared with userspace */ #define NOTE_DELETE 0x0001 /* vnode was removed */ #define NOTE_WRITE 0x0002 /* data contents changed */ #define NOTE_EXTEND 0x0004 /* size increased */ #define NOTE_ATTRIB 0x0008 /* attributes changed */ #define NOTE_LINK 0x0010 /* link count changed */ #define NOTE_RENAME 0x0020 /* vnode was renamed */ #define NOTE_REVOKE 0x0040 /* vnode access was revoked */ #define NOTE_OPEN 0x0080 /* vnode was opened */ #define NOTE_CLOSE 0x0100 /* file closed, fd did not allowed write */ #define NOTE_CLOSE_WRITE 0x0200 /* file closed, fd did allowed write */ #define NOTE_READ 0x0400 /* file was read */ /* * data/hint flags for EVFILT_PROC and EVFILT_PROCDESC, shared with userspace */ #define NOTE_EXIT 0x80000000 /* process exited */ #define NOTE_FORK 0x40000000 /* process forked */ #define NOTE_EXEC 0x20000000 /* process exec'd */ #define NOTE_PCTRLMASK 0xf0000000 /* mask for hint bits */ #define NOTE_PDATAMASK 0x000fffff /* mask for pid */ /* additional flags for EVFILT_PROC */ #define NOTE_TRACK 0x00000001 /* follow across forks */ #define NOTE_TRACKERR 0x00000002 /* could not track child */ #define NOTE_CHILD 0x00000004 /* am a child process */ /* additional flags for EVFILT_TIMER */ #define NOTE_SECONDS 0x00000001 /* data is seconds */ #define NOTE_MSECONDS 0x00000002 /* data is milliseconds */ #define NOTE_USECONDS 0x00000004 /* data is microseconds */ #define NOTE_NSECONDS 0x00000008 /* data is nanoseconds */ #define NOTE_ABSTIME 0x00000010 /* timeout is absolute */ struct knote; SLIST_HEAD(klist, knote); struct kqueue; TAILQ_HEAD(kqlist, kqueue); struct knlist { struct klist kl_list; void (*kl_lock)(void *); /* lock function */ void (*kl_unlock)(void *); - void (*kl_assert_locked)(void *); - void (*kl_assert_unlocked)(void *); + void (*kl_assert_lock)(void *, int); void *kl_lockarg; /* argument passed to lock functions */ int kl_autodestroy; }; #ifdef _KERNEL /* * Flags for knote call */ #define KNF_LISTLOCKED 0x0001 /* knlist is locked */ #define KNF_NOKQLOCK 0x0002 /* do not keep KQ_LOCK */ #define KNOTE(list, hint, flags) knote(list, hint, flags) #define KNOTE_LOCKED(list, hint) knote(list, hint, KNF_LISTLOCKED) #define KNOTE_UNLOCKED(list, hint) knote(list, hint, 0) #define KNLIST_EMPTY(list) SLIST_EMPTY(&(list)->kl_list) /* * Flag indicating hint is a signal. Used by EVFILT_SIGNAL, and also * shared by EVFILT_PROC (all knotes attached to p->p_klist) */ #define NOTE_SIGNAL 0x08000000 /* * Hint values for the optional f_touch event filter. If f_touch is not set * to NULL and f_isfd is zero the f_touch filter will be called with the type * argument set to EVENT_REGISTER during a kevent() system call. It is also * called under the same conditions with the type argument set to EVENT_PROCESS * when the event has been triggered. */ #define EVENT_REGISTER 1 #define EVENT_PROCESS 2 struct filterops { int f_isfd; /* true if ident == filedescriptor */ int (*f_attach)(struct knote *kn); void (*f_detach)(struct knote *kn); int (*f_event)(struct knote *kn, long hint); void (*f_touch)(struct knote *kn, struct kevent *kev, u_long type); }; /* * An in-flux knote cannot be dropped from its kq while the kq is * unlocked. If the KN_SCAN flag is not set, a thread can only set * kn_influx when it is exclusive owner of the knote state, and can * modify kn_status as if it had the KQ lock. KN_SCAN must not be set * on a knote which is already in flux. * * kn_sfflags, kn_sdata, and kn_kevent are protected by the knlist lock. */ struct knote { SLIST_ENTRY(knote) kn_link; /* for kq */ SLIST_ENTRY(knote) kn_selnext; /* for struct selinfo */ struct knlist *kn_knlist; /* f_attach populated */ TAILQ_ENTRY(knote) kn_tqe; struct kqueue *kn_kq; /* which queue we are on */ struct kevent kn_kevent; void *kn_hook; int kn_hookid; int kn_status; /* protected by kq lock */ #define KN_ACTIVE 0x01 /* event has been triggered */ #define KN_QUEUED 0x02 /* event is on queue */ #define KN_DISABLED 0x04 /* event is disabled */ #define KN_DETACHED 0x08 /* knote is detached */ #define KN_MARKER 0x20 /* ignore this knote */ #define KN_KQUEUE 0x40 /* this knote belongs to a kq */ #define KN_SCAN 0x100 /* flux set in kqueue_scan() */ int kn_influx; int kn_sfflags; /* saved filter flags */ int64_t kn_sdata; /* saved data field */ union { struct file *p_fp; /* file data pointer */ struct proc *p_proc; /* proc pointer */ struct kaiocb *p_aio; /* AIO job pointer */ struct aioliojob *p_lio; /* LIO job pointer */ void *p_v; /* generic other pointer */ } kn_ptr; struct filterops *kn_fop; #define kn_id kn_kevent.ident #define kn_filter kn_kevent.filter #define kn_flags kn_kevent.flags #define kn_fflags kn_kevent.fflags #define kn_data kn_kevent.data #define kn_fp kn_ptr.p_fp }; struct kevent_copyops { void *arg; int (*k_copyout)(void *arg, struct kevent *kevp, int count); int (*k_copyin)(void *arg, struct kevent *kevp, int count); size_t kevent_size; }; struct thread; struct proc; struct knlist; struct mtx; struct rwlock; void knote(struct knlist *list, long hint, int lockflags); void knote_fork(struct knlist *list, int pid); struct knlist *knlist_alloc(struct mtx *lock); void knlist_detach(struct knlist *knl); void knlist_add(struct knlist *knl, struct knote *kn, int islocked); void knlist_remove(struct knlist *knl, struct knote *kn, int islocked); int knlist_empty(struct knlist *knl); void knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), - void (*kl_unlock)(void *), void (*kl_assert_locked)(void *), - void (*kl_assert_unlocked)(void *)); + void (*kl_unlock)(void *), void (*kl_assert_lock)(void *, int)); void knlist_init_mtx(struct knlist *knl, struct mtx *lock); void knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock); void knlist_destroy(struct knlist *knl); void knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn); #define knlist_clear(knl, islocked) \ knlist_cleardel((knl), NULL, (islocked), 0) #define knlist_delete(knl, td, islocked) \ knlist_cleardel((knl), (td), (islocked), 1) void knote_fdclose(struct thread *p, int fd); int kqfd_register(int fd, struct kevent *kev, struct thread *p, int mflag); int kqueue_add_filteropts(int filt, struct filterops *filtops); int kqueue_del_filteropts(int filt); #else /* !_KERNEL */ #include struct timespec; __BEGIN_DECLS int kqueue(void); int kevent(int kq, const struct kevent *changelist, int nchanges, struct kevent *eventlist, int nevents, const struct timespec *timeout); __END_DECLS #endif /* !_KERNEL */ #endif /* !_SYS_EVENT_H_ */