diff options
Diffstat (limited to 'sys/kern/vfs_syscalls.c')
| -rw-r--r-- | sys/kern/vfs_syscalls.c | 134 |
1 files changed, 134 insertions, 0 deletions
diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c index ec5ad06..e886079 100644 --- a/sys/kern/vfs_syscalls.c +++ b/sys/kern/vfs_syscalls.c @@ -86,6 +86,8 @@ __FBSDID("$FreeBSD$"); #include <vm/vm_page.h> #include <vm/uma.h> +static MALLOC_DEFINE(M_FADVISE, "fadvise", "posix_fadvise(2) information"); + SDT_PROVIDER_DEFINE(vfs); SDT_PROBE_DEFINE(vfs, , stat, mode, mode); SDT_PROBE_ARGTYPE(vfs, , stat, mode, 0, "char *"); @@ -4845,3 +4847,135 @@ sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap) return (kern_posix_fallocate(td, uap->fd, uap->offset, uap->len)); } + +/* + * Unlike madvise(2), we do not make a best effort to remember every + * possible caching hint. Instead, we remember the last setting with + * the exception that we will allow POSIX_FADV_NORMAL to adjust the + * region of any current setting. + */ +int +sys_posix_fadvise(struct thread *td, struct posix_fadvise_args *uap) +{ + struct fadvise_info *fa, *new; + struct file *fp; + struct vnode *vp; + off_t end; + int error; + + if (uap->offset < 0 || uap->len < 0 || + uap->offset > OFF_MAX - uap->len) + return (EINVAL); + switch (uap->advice) { + case POSIX_FADV_SEQUENTIAL: + case POSIX_FADV_RANDOM: + case POSIX_FADV_NOREUSE: + new = malloc(sizeof(*fa), M_FADVISE, M_WAITOK); + break; + case POSIX_FADV_NORMAL: + case POSIX_FADV_WILLNEED: + case POSIX_FADV_DONTNEED: + new = NULL; + break; + default: + return (EINVAL); + } + /* XXX: CAP_POSIX_FADVISE? */ + error = fget(td, uap->fd, 0, &fp); + if (error != 0) + goto out; + + switch (fp->f_type) { + case DTYPE_VNODE: + break; + case DTYPE_PIPE: + case DTYPE_FIFO: + error = ESPIPE; + goto out; + default: + error = ENODEV; + goto out; + } + vp = fp->f_vnode; + if (vp->v_type != VREG) { + error = ENODEV; + goto out; + } + if (uap->len == 0) + end = OFF_MAX; + else + end = uap->offset + uap->len - 1; + switch (uap->advice) { + case POSIX_FADV_SEQUENTIAL: + case POSIX_FADV_RANDOM: + case POSIX_FADV_NOREUSE: + /* + * Try to merge any existing non-standard region with + * this new region if possible, otherwise create a new + * non-standard region for this request. + */ + mtx_pool_lock(mtxpool_sleep, fp); + fa = fp->f_advice; + if (fa != NULL && fa->fa_advice == uap->advice && + ((fa->fa_start <= end && fa->fa_end >= uap->offset) || + (end != OFF_MAX && fa->fa_start == end + 1) || + (fa->fa_end != OFF_MAX && fa->fa_end + 1 == uap->offset))) { + if (uap->offset < fa->fa_start) + fa->fa_start = uap->offset; + if (end > fa->fa_end) + fa->fa_end = end; + } else { + new->fa_advice = uap->advice; + new->fa_start = uap->offset; + new->fa_end = end; + fp->f_advice = new; + new = fa; + } + mtx_pool_unlock(mtxpool_sleep, fp); + break; + case POSIX_FADV_NORMAL: + /* + * If a the "normal" region overlaps with an existing + * non-standard region, trim or remove the + * non-standard region. + */ + mtx_pool_lock(mtxpool_sleep, fp); + fa = fp->f_advice; + if (fa != NULL) { + if (uap->offset <= fa->fa_start && + end >= fa->fa_end) { + new = fa; + fp->f_advice = NULL; + } else if (uap->offset <= fa->fa_start && + end >= fa->fa_start) + fa->fa_start = end + 1; + else if (uap->offset <= fa->fa_end && + end >= fa->fa_end) + fa->fa_end = uap->offset - 1; + else if (uap->offset >= fa->fa_start && + end <= fa->fa_end) { + /* + * If the "normal" region is a middle + * portion of the existing + * non-standard region, just remove + * the whole thing rather than picking + * one side or the other to + * preserve. + */ + new = fa; + fp->f_advice = NULL; + } + } + mtx_pool_unlock(mtxpool_sleep, fp); + break; + case POSIX_FADV_WILLNEED: + case POSIX_FADV_DONTNEED: + error = VOP_ADVISE(vp, uap->offset, end, uap->advice); + break; + } +out: + if (fp != NULL) + fdrop(fp, td); + free(new, M_FADVISE); + return (error); +} |
