Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
[sfrench/cifs-2.6.git] / ipc / shm.c
index 5bb617f6306e8df814398505393b15e2b0c3c0ee..0852f206d8951cb4f53071b36e159ae0332e5093 100644 (file)
--- a/ipc/shm.c
+++ b/ipc/shm.c
 #include <linux/seq_file.h>
 #include <linux/mutex.h>
 #include <linux/nsproxy.h>
+#include <linux/mount.h>
 
 #include <asm/uaccess.h>
 
 #include "util.h"
 
+struct shm_file_data {
+       int id;
+       struct ipc_namespace *ns;
+       struct file *file;
+       const struct vm_operations_struct *vm_ops;
+};
+
+#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
+
 static const struct file_operations shm_file_operations;
 static struct vm_operations_struct shm_vm_ops;
 
@@ -60,8 +70,8 @@ static struct ipc_ids init_shm_ids;
 
 static int newseg (struct ipc_namespace *ns, key_t key,
                int shmflg, size_t size);
-static void shm_open (struct vm_area_struct *shmd);
-static void shm_close (struct vm_area_struct *shmd);
+static void shm_open(struct vm_area_struct *vma);
+static void shm_close(struct vm_area_struct *vma);
 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
 #ifdef CONFIG_PROC_FS
 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
@@ -150,11 +160,14 @@ static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 
 
 
-static inline void shm_inc(struct ipc_namespace *ns, int id)
+/* This is called by fork, once for every shm attach. */
+static void shm_open(struct vm_area_struct *vma)
 {
+       struct file *file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
        struct shmid_kernel *shp;
 
-       shp = shm_lock(ns, id);
+       shp = shm_lock(sfd->ns, sfd->id);
        BUG_ON(!shp);
        shp->shm_atim = get_seconds();
        shp->shm_lprid = current->tgid;
@@ -162,15 +175,6 @@ static inline void shm_inc(struct ipc_namespace *ns, int id)
        shm_unlock(shp);
 }
 
-#define shm_file_ns(file) (*((struct ipc_namespace **)&(file)->private_data))
-
-/* This is called by fork, once for every shm attach. */
-static void shm_open(struct vm_area_struct *shmd)
-{
-       shm_inc(shm_file_ns(shmd->vm_file),
-                       shmd->vm_file->f_path.dentry->d_inode->i_ino);
-}
-
 /*
  * shm_destroy - free the struct shmid_kernel
  *
@@ -195,23 +199,21 @@ static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 }
 
 /*
- * remove the attach descriptor shmd.
+ * remove the attach descriptor vma.
  * free memory for segment if it is marked destroyed.
  * The descriptor has already been removed from the current->mm->mmap list
  * and will later be kfree()d.
  */
-static void shm_close (struct vm_area_struct *shmd)
+static void shm_close(struct vm_area_struct *vma)
 {
-       struct file * file = shmd->vm_file;
-       int id = file->f_path.dentry->d_inode->i_ino;
+       struct file * file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
        struct shmid_kernel *shp;
-       struct ipc_namespace *ns;
-
-       ns = shm_file_ns(file);
+       struct ipc_namespace *ns = sfd->ns;
 
        mutex_lock(&shm_ids(ns).mutex);
        /* remove from the list of attaches of the shm segment */
-       shp = shm_lock(ns, id);
+       shp = shm_lock(ns, sfd->id);
        BUG_ON(!shp);
        shp->shm_lprid = current->tgid;
        shp->shm_dtim = get_seconds();
@@ -224,46 +226,113 @@ static void shm_close (struct vm_area_struct *shmd)
        mutex_unlock(&shm_ids(ns).mutex);
 }
 
+static struct page *shm_nopage(struct vm_area_struct *vma,
+                              unsigned long address, int *type)
+{
+       struct file *file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
+
+       return sfd->vm_ops->nopage(vma, address, type);
+}
+
+#ifdef CONFIG_NUMA
+int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
+{
+       struct file *file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
+       int err = 0;
+       if (sfd->vm_ops->set_policy)
+               err = sfd->vm_ops->set_policy(vma, new);
+       return err;
+}
+
+struct mempolicy *shm_get_policy(struct vm_area_struct *vma, unsigned long addr)
+{
+       struct file *file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
+       struct mempolicy *pol = NULL;
+
+       if (sfd->vm_ops->get_policy)
+               pol = sfd->vm_ops->get_policy(vma, addr);
+       else if (vma->vm_policy)
+               pol = vma->vm_policy;
+       else
+               pol = current->mempolicy;
+       return pol;
+}
+#endif
+
 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 {
+       struct shm_file_data *sfd = shm_file_data(file);
        int ret;
 
-       ret = shmem_mmap(file, vma);
-       if (ret == 0) {
-               vma->vm_ops = &shm_vm_ops;
-               if (!(vma->vm_flags & VM_WRITE))
-                       vma->vm_flags &= ~VM_MAYWRITE;
-               shm_inc(shm_file_ns(file), file->f_path.dentry->d_inode->i_ino);
-       }
+       ret = sfd->file->f_op->mmap(sfd->file, vma);
+       if (ret != 0)
+               return ret;
+       sfd->vm_ops = vma->vm_ops;
+       vma->vm_ops = &shm_vm_ops;
+       shm_open(vma);
 
        return ret;
 }
 
 static int shm_release(struct inode *ino, struct file *file)
 {
-       struct ipc_namespace *ns;
+       struct shm_file_data *sfd = shm_file_data(file);
 
-       ns = shm_file_ns(file);
-       put_ipc_ns(ns);
-       shm_file_ns(file) = NULL;
+       put_ipc_ns(sfd->ns);
+       shm_file_data(file) = NULL;
+       kfree(sfd);
        return 0;
 }
 
+static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
+{
+       int (*fsync) (struct file *, struct dentry *, int datasync);
+       struct shm_file_data *sfd = shm_file_data(file);
+       int ret = -EINVAL;
+
+       fsync = sfd->file->f_op->fsync;
+       if (fsync)
+               ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
+       return ret;
+}
+
+static unsigned long shm_get_unmapped_area(struct file *file,
+       unsigned long addr, unsigned long len, unsigned long pgoff,
+       unsigned long flags)
+{
+       struct shm_file_data *sfd = shm_file_data(file);
+       return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
+}
+
+int is_file_shm_hugepages(struct file *file)
+{
+       int ret = 0;
+
+       if (file->f_op == &shm_file_operations) {
+               struct shm_file_data *sfd;
+               sfd = shm_file_data(file);
+               ret = is_file_hugepages(sfd->file);
+       }
+       return ret;
+}
+
 static const struct file_operations shm_file_operations = {
        .mmap           = shm_mmap,
+       .fsync          = shm_fsync,
        .release        = shm_release,
-#ifndef CONFIG_MMU
-       .get_unmapped_area = shmem_get_unmapped_area,
-#endif
+       .get_unmapped_area      = shm_get_unmapped_area,
 };
 
 static struct vm_operations_struct shm_vm_ops = {
        .open   = shm_open,     /* callback for a new vm-area open */
        .close  = shm_close,    /* callback for when the vm-area is released */
-       .nopage = shmem_nopage,
-#if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
-       .set_policy = shmem_set_policy,
-       .get_policy = shmem_get_policy,
+       .nopage = shm_nopage,
+#if defined(CONFIG_NUMA)
+       .set_policy = shm_set_policy,
+       .get_policy = shm_get_policy,
 #endif
 };
 
@@ -297,9 +366,10 @@ static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
                return error;
        }
 
+       sprintf (name, "SYSV%08x", key);
        if (shmflg & SHM_HUGETLB) {
-               /* hugetlb_zero_setup takes care of mlock user accounting */
-               file = hugetlb_zero_setup(size);
+               /* hugetlb_file_setup takes care of mlock user accounting */
+               file = hugetlb_file_setup(name, size);
                shp->mlock_user = current->user;
        } else {
                int acctflag = VM_ACCOUNT;
@@ -310,7 +380,6 @@ static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
                if  ((shmflg & SHM_NORESERVE) &&
                                sysctl_overcommit_memory != OVERCOMMIT_NEVER)
                        acctflag = 0;
-               sprintf (name, "SYSV%08x", key);
                file = shmem_file_setup(name, size, acctflag);
        }
        error = PTR_ERR(file);
@@ -330,13 +399,11 @@ static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
        shp->shm_nattch = 0;
        shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
        shp->shm_file = file;
-       file->f_path.dentry->d_inode->i_ino = shp->id;
-
-       shm_file_ns(file) = get_ipc_ns(ns);
-
-       /* Hugetlb ops would have already been assigned. */
-       if (!(shmflg & SHM_HUGETLB))
-               file->f_op = &shm_file_operations;
+       /*
+        * shmid gets reported as "inode#" in /proc/pid/maps.
+        * proc-ps tools use this. Changing this will break them.
+        */
+       file->f_dentry->d_inode->i_ino = shp->id;
 
        ns->shm_tot += numpages;
        shm_unlock(shp);
@@ -607,10 +674,7 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                tbuf.shm_ctime  = shp->shm_ctim;
                tbuf.shm_cpid   = shp->shm_cprid;
                tbuf.shm_lpid   = shp->shm_lprid;
-               if (!is_file_hugepages(shp->shm_file))
-                       tbuf.shm_nattch = shp->shm_nattch;
-               else
-                       tbuf.shm_nattch = file_count(shp->shm_file) - 1;
+               tbuf.shm_nattch = shp->shm_nattch;
                shm_unlock(shp);
                if(copy_shmid_to_user (buf, &tbuf, version))
                        err = -EFAULT;
@@ -779,13 +843,16 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        unsigned long flags;
        unsigned long prot;
        int acc_mode;
-       void *user_addr;
+       unsigned long user_addr;
        struct ipc_namespace *ns;
+       struct shm_file_data *sfd;
+       struct path path;
+       mode_t f_mode;
 
-       if (shmid < 0) {
-               err = -EINVAL;
+       err = -EINVAL;
+       if (shmid < 0)
                goto out;
-       else if ((addr = (ulong)shmaddr)) {
+       else if ((addr = (ulong)shmaddr)) {
                if (addr & (SHMLBA-1)) {
                        if (shmflg & SHM_RND)
                                addr &= ~(SHMLBA-1);       /* round down */
@@ -793,12 +860,12 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 #ifndef __ARCH_FORCE_SHMLBA
                                if (addr & ~PAGE_MASK)
 #endif
-                                       return -EINVAL;
+                                       goto out;
                }
                flags = MAP_SHARED | MAP_FIXED;
        } else {
                if ((shmflg & SHM_REMAP))
-                       return -EINVAL;
+                       goto out;
 
                flags = MAP_SHARED;
        }
@@ -806,9 +873,11 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        if (shmflg & SHM_RDONLY) {
                prot = PROT_READ;
                acc_mode = S_IRUGO;
+               f_mode = FMODE_READ;
        } else {
                prot = PROT_READ | PROT_WRITE;
                acc_mode = S_IRUGO | S_IWUGO;
+               f_mode = FMODE_READ | FMODE_WRITE;
        }
        if (shmflg & SHM_EXEC) {
                prot |= PROT_EXEC;
@@ -821,35 +890,50 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
         */
        ns = current->nsproxy->ipc_ns;
        shp = shm_lock(ns, shmid);
-       if(shp == NULL) {
-               err = -EINVAL;
+       if(shp == NULL)
                goto out;
-       }
+
        err = shm_checkid(ns, shp,shmid);
-       if (err) {
-               shm_unlock(shp);
-               goto out;
-       }
-       if (ipcperms(&shp->shm_perm, acc_mode)) {
-               shm_unlock(shp);
-               err = -EACCES;
-               goto out;
-       }
+       if (err)
+               goto out_unlock;
+
+       err = -EACCES;
+       if (ipcperms(&shp->shm_perm, acc_mode))
+               goto out_unlock;
 
        err = security_shm_shmat(shp, shmaddr, shmflg);
-       if (err) {
-               shm_unlock(shp);
-               return err;
-       }
-               
-       file = shp->shm_file;
-       size = i_size_read(file->f_path.dentry->d_inode);
+       if (err)
+               goto out_unlock;
+
+       path.dentry = dget(shp->shm_file->f_path.dentry);
+       path.mnt    = mntget(shp->shm_file->f_path.mnt);
        shp->shm_nattch++;
+       size = i_size_read(path.dentry->d_inode);
        shm_unlock(shp);
 
+       err = -ENOMEM;
+       sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
+       if (!sfd)
+               goto out_put_path;
+
+       err = -ENOMEM;
+       file = get_empty_filp();
+       if (!file)
+               goto out_free;
+
+       file->f_op = &shm_file_operations;
+       file->private_data = sfd;
+       file->f_path = path;
+       file->f_mapping = shp->shm_file->f_mapping;
+       file->f_mode = f_mode;
+       sfd->id = shp->id;
+       sfd->ns = get_ipc_ns(ns);
+       sfd->file = shp->shm_file;
+       sfd->vm_ops = NULL;
+
        down_write(&current->mm->mmap_sem);
        if (addr && !(shmflg & SHM_REMAP)) {
-               user_addr = ERR_PTR(-EINVAL);
+               err = -EINVAL;
                if (find_vma_intersection(current->mm, addr, addr + size))
                        goto invalid;
                /*
@@ -861,11 +945,17 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
                        goto invalid;
        }
                
-       user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
-
+       user_addr = do_mmap (file, addr, size, prot, flags, 0);
+       *raddr = user_addr;
+       err = 0;
+       if (IS_ERR_VALUE(user_addr))
+               err = (long)user_addr;
 invalid:
        up_write(&current->mm->mmap_sem);
 
+       fput(file);
+
+out_nattch:
        mutex_lock(&shm_ids(ns).mutex);
        shp = shm_lock(ns, shmid);
        BUG_ON(!shp);
@@ -877,12 +967,19 @@ invalid:
                shm_unlock(shp);
        mutex_unlock(&shm_ids(ns).mutex);
 
-       *raddr = (unsigned long) user_addr;
-       err = 0;
-       if (IS_ERR(user_addr))
-               err = PTR_ERR(user_addr);
 out:
        return err;
+
+out_unlock:
+       shm_unlock(shp);
+       goto out;
+
+out_free:
+       kfree(sfd);
+out_put_path:
+       dput(path.dentry);
+       mntput(path.mnt);
+       goto out_nattch;
 }
 
 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
@@ -944,7 +1041,7 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
                 * a fragment created by mprotect() and/or munmap(), or it
                 * otherwise it starts at this address with no hassles.
                 */
-               if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
+               if ((vma->vm_ops == &shm_vm_ops) &&
                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 
 
@@ -973,7 +1070,7 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
                next = vma->vm_next;
 
                /* finding a matching vma now does not alter retval */
-               if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
+               if ((vma->vm_ops == &shm_vm_ops) &&
                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 
                        do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
@@ -1004,7 +1101,7 @@ static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
                          shp->shm_segsz,
                          shp->shm_cprid,
                          shp->shm_lprid,
-                         is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
+                         shp->shm_nattch,
                          shp->shm_perm.uid,
                          shp->shm_perm.gid,
                          shp->shm_perm.cuid,