1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
40 #ifndef AUTOCONF_INCLUDED
41 #include <linux/config.h>
43 #include <linux/kernel.h>
45 #include <linux/string.h>
46 #include <linux/stat.h>
47 #include <linux/errno.h>
48 #include <linux/smp_lock.h>
49 #include <linux/unistd.h>
50 #include <linux/version.h>
51 #include <asm/system.h>
52 #include <asm/uaccess.h>
55 #include <linux/stat.h>
56 #include <asm/uaccess.h>
58 #include <linux/pagemap.h>
59 #include <linux/smp_lock.h>
61 #define DEBUG_SUBSYSTEM S_LLITE
63 //#include <lustre_mdc.h>
64 #include <lustre_lite.h>
65 #include "llite_internal.h"
66 #include <linux/lustre_compat25.h>
68 #define VMA_DEBUG(vma, fmt, arg...) \
69 CDEBUG(D_MMAP, "vma(%p) start(%ld) end(%ld) pgoff(%ld) inode(%p) " \
70 "ino(%lu) iname(%s): " fmt, vma, vma->vm_start, vma->vm_end, \
71 vma->vm_pgoff, vma->vm_file->f_dentry->d_inode, \
72 vma->vm_file->f_dentry->d_inode->i_ino, \
73 vma->vm_file->f_dentry->d_iname, ## arg); \
75 struct page *ll_nopage(struct vm_area_struct *vma, unsigned long address,
78 static struct vm_operations_struct ll_file_vm_ops;
80 void policy_from_vma(ldlm_policy_data_t *policy,
81 struct vm_area_struct *vma, unsigned long addr,
84 policy->l_extent.start = ((addr - vma->vm_start) & CFS_PAGE_MASK) +
85 (vma->vm_pgoff << CFS_PAGE_SHIFT);
86 policy->l_extent.end = (policy->l_extent.start + count - 1) |
90 struct vm_area_struct * our_vma(unsigned long addr, size_t count)
92 struct mm_struct *mm = current->mm;
93 struct vm_area_struct *vma, *ret = NULL;
96 /* No MM (e.g. NFS)? No vmas too. */
100 spin_lock(&mm->page_table_lock);
101 for(vma = find_vma(mm, addr);
102 vma != NULL && vma->vm_start < (addr + count); vma = vma->vm_next) {
103 if (vma->vm_ops && vma->vm_ops == &ll_file_vm_ops &&
104 vma->vm_flags & VM_SHARED) {
109 spin_unlock(&mm->page_table_lock);
114 * API independent part for page fault initialization.
115 * \param vma - virtual memory area addressed to page fault
116 * \param env - corespondent lu_env to processing
117 * \param nest - nested level
118 * \param index - page index corespondent to fault.
119 * \parm ra_flags - vma readahead flags.
121 * \return allocated and initialized env for fault operation.
122 * \retval EINVAL if env can't allocated
123 * \return other error codes from cl_io_init.
125 struct cl_io *ll_fault_io_init(struct vm_area_struct *vma,
126 struct lu_env **env_ret,
127 struct cl_env_nest *nest,
128 pgoff_t index, unsigned long *ra_flags)
130 struct file *file = vma->vm_file;
131 struct inode *inode = file->f_dentry->d_inode;
132 const unsigned long writable = VM_SHARED|VM_WRITE;
134 struct cl_fault_io *fio;
139 if (ll_file_nolock(file))
140 RETURN(ERR_PTR(-EOPNOTSUPP));
143 * page fault can be called when lustre IO is
144 * already active for the current thread, e.g., when doing read/write
145 * against user level buffer mapped from Lustre buffer. To avoid
146 * stomping on existing context, optionally force an allocation of a new
149 env = cl_env_nested_get(nest);
151 RETURN(ERR_PTR(-EINVAL));
155 io = ccc_env_thread_io(env);
156 io->ci_obj = ll_i2info(inode)->lli_clob;
157 LASSERT(io->ci_obj != NULL);
159 fio = &io->u.ci_fault;
160 fio->ft_index = index;
161 fio->ft_writable = (vma->vm_flags&writable) == writable;
162 fio->ft_executable = vma->vm_flags&VM_EXEC;
165 * disable VM_SEQ_READ and use VM_RAND_READ to make sure that
166 * the kernel will not read other pages not covered by ldlm in
167 * filemap_nopage. we do our readahead in ll_readpage.
169 *ra_flags = vma->vm_flags & (VM_RAND_READ|VM_SEQ_READ);
170 vma->vm_flags &= ~VM_SEQ_READ;
171 vma->vm_flags |= VM_RAND_READ;
173 CDEBUG(D_INFO, "vm_flags: %lx (%lu %d %d)\n", vma->vm_flags,
174 fio->ft_index, fio->ft_writable, fio->ft_executable);
176 if (cl_io_init(env, io, CIT_FAULT, io->ci_obj) == 0) {
177 struct ccc_io *cio = ccc_env_io(env);
178 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
180 LASSERT(cio->cui_cl.cis_io == io);
182 /* mmap lock must be MANDATORY
183 * it has to cache pages. */
184 io->ci_lockreq = CILR_MANDATORY;
192 #ifndef HAVE_VM_OP_FAULT
194 * Lustre implementation of a vm_operations_struct::nopage() method, called by
195 * VM to server page fault (both in kernel and user space).
197 * This function sets up CIT_FAULT cl_io that does the job.
199 * \param vma - is virtiual area struct related to page fault
200 * \param address - address when hit fault
201 * \param type - of fault
203 * \return allocated and filled _unlocked_ page for address
204 * \retval NOPAGE_SIGBUS if page not exist on this address
205 * \retval NOPAGE_OOM not have memory for allocate new page
207 struct page *ll_nopage(struct vm_area_struct *vma, unsigned long address,
211 struct cl_env_nest nest;
213 struct page *page = NOPAGE_SIGBUS;
214 struct vvp_io *vio = NULL;
215 unsigned long ra_flags;
220 pg_offset = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
221 io = ll_fault_io_init(vma, &env, &nest, pg_offset, &ra_flags);
223 return NOPAGE_SIGBUS;
225 result = io->ci_result;
229 vio = vvp_env_io(env);
230 vio->u.fault.ft_vma = vma;
231 vio->u.fault.nopage.ft_address = address;
232 vio->u.fault.nopage.ft_type = type;
234 result = cl_io_loop(env, io);
238 LASSERT(io->u.ci_fault.ft_page != NULL);
239 page = vio->u.fault.ft_vmpage;
241 if (result == -ENOMEM)
245 vma->vm_flags &= ~VM_RAND_READ;
246 vma->vm_flags |= ra_flags;
249 cl_env_nested_put(&nest, env);
255 * Lustre implementation of a vm_operations_struct::fault() method, called by
256 * VM to server page fault (both in kernel and user space).
258 * \param vma - is virtiual area struct related to page fault
259 * \param vmf - structure which describe type and address where hit fault
261 * \return allocated and filled _locked_ page for address
262 * \retval VM_FAULT_ERROR on general error
263 * \retval NOPAGE_OOM not have memory for allocate new page
265 int ll_fault0(struct vm_area_struct *vma, struct vm_fault *vmf)
269 struct vvp_io *vio = NULL;
270 unsigned long ra_flags;
271 struct cl_env_nest nest;
276 io = ll_fault_io_init(vma, &env, &nest, vmf->pgoff, &ra_flags);
278 RETURN(VM_FAULT_ERROR);
280 result = io->ci_result;
284 vio = vvp_env_io(env);
285 vio->u.fault.ft_vma = vma;
286 vio->u.fault.ft_vmpage = NULL;
287 vio->u.fault.fault.ft_vmf = vmf;
289 result = cl_io_loop(env, io);
290 fault_ret = vio->u.fault.fault.ft_flags;
293 if ((result != 0) && !(fault_ret & VM_FAULT_RETRY))
294 fault_ret |= VM_FAULT_ERROR;
296 vma->vm_flags |= ra_flags;
299 cl_env_nested_put(&nest, env);
304 int ll_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
307 bool printed = false;
311 result = ll_fault0(vma, vmf);
312 LASSERT(!(result & VM_FAULT_LOCKED));
314 struct page *vmpage = vmf->page;
316 /* check if this page has been truncated */
318 if (unlikely(vmpage->mapping == NULL)) { /* unlucky */
320 page_cache_release(vmpage);
323 if (!printed && ++count > 16) {
324 CWARN("the page is under heavy contention,"
325 "maybe your app(%s) needs revising :-)\n",
333 result |= VM_FAULT_LOCKED;
340 * To avoid cancel the locks covering mmapped region for lock cache pressure,
341 * we track the mapped vma count in ccc_object::cob_mmap_cnt.
343 static void ll_vm_open(struct vm_area_struct * vma)
345 struct inode *inode = vma->vm_file->f_dentry->d_inode;
346 struct ccc_object *vob = cl_inode2ccc(inode);
349 LASSERT(vma->vm_file);
350 LASSERT(cfs_atomic_read(&vob->cob_mmap_cnt) >= 0);
351 cfs_atomic_inc(&vob->cob_mmap_cnt);
356 * Dual to ll_vm_open().
358 static void ll_vm_close(struct vm_area_struct *vma)
360 struct inode *inode = vma->vm_file->f_dentry->d_inode;
361 struct ccc_object *vob = cl_inode2ccc(inode);
364 LASSERT(vma->vm_file);
365 cfs_atomic_dec(&vob->cob_mmap_cnt);
366 LASSERT(cfs_atomic_read(&vob->cob_mmap_cnt) >= 0);
370 #ifndef HAVE_VM_OP_FAULT
371 #ifndef HAVE_FILEMAP_POPULATE
372 static int (*filemap_populate)(struct vm_area_struct * area, unsigned long address, unsigned long len, pgprot_t prot, unsigned long pgoff, int nonblock);
374 static int ll_populate(struct vm_area_struct *area, unsigned long address,
375 unsigned long len, pgprot_t prot, unsigned long pgoff,
381 /* always set nonblock as true to avoid page read ahead */
382 rc = filemap_populate(area, address, len, prot, pgoff, 1);
387 /* return the user space pointer that maps to a file offset via a vma */
388 static inline unsigned long file_to_user(struct vm_area_struct *vma, __u64 byte)
390 return vma->vm_start + (byte - ((__u64)vma->vm_pgoff << CFS_PAGE_SHIFT));
394 /* XXX put nice comment here. talk about __free_pte -> dirty pages and
395 * nopage's reference passing to the pte */
396 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last)
401 LASSERTF(last > first, "last "LPU64" first "LPU64"\n", last, first);
402 if (mapping_mapped(mapping)) {
404 unmap_mapping_range(mapping, first + CFS_PAGE_SIZE - 1,
405 last - first + 1, 0);
411 static struct vm_operations_struct ll_file_vm_ops = {
412 #ifndef HAVE_VM_OP_FAULT
414 .populate = ll_populate,
420 .close = ll_vm_close,
423 int ll_file_mmap(struct file *file, struct vm_area_struct * vma)
425 struct inode *inode = file->f_dentry->d_inode;
429 if (ll_file_nolock(file))
432 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_MAP, 1);
433 rc = generic_file_mmap(file, vma);
435 #if !defined(HAVE_FILEMAP_POPULATE) && !defined(HAVE_VM_OP_FAULT)
436 if (!filemap_populate)
437 filemap_populate = vma->vm_ops->populate;
439 vma->vm_ops = &ll_file_vm_ops;
440 vma->vm_ops->open(vma);
441 /* update the inode's size and mtime */
442 rc = cl_glimpse_size(inode);