Back to home page

LXR

 
 

    


0001 /*
0002  *  linux/mm/filemap.c
0003  *
0004  * Copyright (C) 1994-1999  Linus Torvalds
0005  */
0006 
0007 /*
0008  * This file handles the generic file mmap semantics used by
0009  * most "normal" filesystems (but you don't /have/ to use this:
0010  * the NFS filesystem used to do this differently, for example)
0011  */
0012 #include <linux/export.h>
0013 #include <linux/compiler.h>
0014 #include <linux/dax.h>
0015 #include <linux/fs.h>
0016 #include <linux/uaccess.h>
0017 #include <linux/capability.h>
0018 #include <linux/kernel_stat.h>
0019 #include <linux/gfp.h>
0020 #include <linux/mm.h>
0021 #include <linux/swap.h>
0022 #include <linux/mman.h>
0023 #include <linux/pagemap.h>
0024 #include <linux/file.h>
0025 #include <linux/uio.h>
0026 #include <linux/hash.h>
0027 #include <linux/writeback.h>
0028 #include <linux/backing-dev.h>
0029 #include <linux/pagevec.h>
0030 #include <linux/blkdev.h>
0031 #include <linux/security.h>
0032 #include <linux/cpuset.h>
0033 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
0034 #include <linux/hugetlb.h>
0035 #include <linux/memcontrol.h>
0036 #include <linux/cleancache.h>
0037 #include <linux/rmap.h>
0038 #include "internal.h"
0039 
0040 #define CREATE_TRACE_POINTS
0041 #include <trace/events/filemap.h>
0042 
0043 /*
0044  * FIXME: remove all knowledge of the buffer layer from the core VM
0045  */
0046 #include <linux/buffer_head.h> /* for try_to_free_buffers */
0047 
0048 #include <asm/mman.h>
0049 
0050 /*
0051  * Shared mappings implemented 30.11.1994. It's not fully working yet,
0052  * though.
0053  *
0054  * Shared mappings now work. 15.8.1995  Bruno.
0055  *
0056  * finished 'unifying' the page and buffer cache and SMP-threaded the
0057  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
0058  *
0059  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
0060  */
0061 
0062 /*
0063  * Lock ordering:
0064  *
0065  *  ->i_mmap_rwsem      (truncate_pagecache)
0066  *    ->private_lock        (__free_pte->__set_page_dirty_buffers)
0067  *      ->swap_lock     (exclusive_swap_page, others)
0068  *        ->mapping->tree_lock
0069  *
0070  *  ->i_mutex
0071  *    ->i_mmap_rwsem        (truncate->unmap_mapping_range)
0072  *
0073  *  ->mmap_sem
0074  *    ->i_mmap_rwsem
0075  *      ->page_table_lock or pte_lock   (various, mainly in memory.c)
0076  *        ->mapping->tree_lock  (arch-dependent flush_dcache_mmap_lock)
0077  *
0078  *  ->mmap_sem
0079  *    ->lock_page       (access_process_vm)
0080  *
0081  *  ->i_mutex           (generic_perform_write)
0082  *    ->mmap_sem        (fault_in_pages_readable->do_page_fault)
0083  *
0084  *  bdi->wb.list_lock
0085  *    sb_lock           (fs/fs-writeback.c)
0086  *    ->mapping->tree_lock  (__sync_single_inode)
0087  *
0088  *  ->i_mmap_rwsem
0089  *    ->anon_vma.lock       (vma_adjust)
0090  *
0091  *  ->anon_vma.lock
0092  *    ->page_table_lock or pte_lock (anon_vma_prepare and various)
0093  *
0094  *  ->page_table_lock or pte_lock
0095  *    ->swap_lock       (try_to_unmap_one)
0096  *    ->private_lock        (try_to_unmap_one)
0097  *    ->tree_lock       (try_to_unmap_one)
0098  *    ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
0099  *    ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
0100  *    ->private_lock        (page_remove_rmap->set_page_dirty)
0101  *    ->tree_lock       (page_remove_rmap->set_page_dirty)
0102  *    bdi.wb->list_lock     (page_remove_rmap->set_page_dirty)
0103  *    ->inode->i_lock       (page_remove_rmap->set_page_dirty)
0104  *    ->memcg->move_lock    (page_remove_rmap->lock_page_memcg)
0105  *    bdi.wb->list_lock     (zap_pte_range->set_page_dirty)
0106  *    ->inode->i_lock       (zap_pte_range->set_page_dirty)
0107  *    ->private_lock        (zap_pte_range->__set_page_dirty_buffers)
0108  *
0109  * ->i_mmap_rwsem
0110  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
0111  */
0112 
0113 static int page_cache_tree_insert(struct address_space *mapping,
0114                   struct page *page, void **shadowp)
0115 {
0116     struct radix_tree_node *node;
0117     void **slot;
0118     int error;
0119 
0120     error = __radix_tree_create(&mapping->page_tree, page->index, 0,
0121                     &node, &slot);
0122     if (error)
0123         return error;
0124     if (*slot) {
0125         void *p;
0126 
0127         p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
0128         if (!radix_tree_exceptional_entry(p))
0129             return -EEXIST;
0130 
0131         mapping->nrexceptional--;
0132         if (!dax_mapping(mapping)) {
0133             if (shadowp)
0134                 *shadowp = p;
0135         } else {
0136             /* DAX can replace empty locked entry with a hole */
0137             WARN_ON_ONCE(p !=
0138                 dax_radix_locked_entry(0, RADIX_DAX_EMPTY));
0139             /* Wakeup waiters for exceptional entry lock */
0140             dax_wake_mapping_entry_waiter(mapping, page->index, p,
0141                               true);
0142         }
0143     }
0144     __radix_tree_replace(&mapping->page_tree, node, slot, page,
0145                  workingset_update_node, mapping);
0146     mapping->nrpages++;
0147     return 0;
0148 }
0149 
0150 static void page_cache_tree_delete(struct address_space *mapping,
0151                    struct page *page, void *shadow)
0152 {
0153     int i, nr;
0154 
0155     /* hugetlb pages are represented by one entry in the radix tree */
0156     nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
0157 
0158     VM_BUG_ON_PAGE(!PageLocked(page), page);
0159     VM_BUG_ON_PAGE(PageTail(page), page);
0160     VM_BUG_ON_PAGE(nr != 1 && shadow, page);
0161 
0162     for (i = 0; i < nr; i++) {
0163         struct radix_tree_node *node;
0164         void **slot;
0165 
0166         __radix_tree_lookup(&mapping->page_tree, page->index + i,
0167                     &node, &slot);
0168 
0169         VM_BUG_ON_PAGE(!node && nr != 1, page);
0170 
0171         radix_tree_clear_tags(&mapping->page_tree, node, slot);
0172         __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
0173                      workingset_update_node, mapping);
0174     }
0175 
0176     if (shadow) {
0177         mapping->nrexceptional += nr;
0178         /*
0179          * Make sure the nrexceptional update is committed before
0180          * the nrpages update so that final truncate racing
0181          * with reclaim does not see both counters 0 at the
0182          * same time and miss a shadow entry.
0183          */
0184         smp_wmb();
0185     }
0186     mapping->nrpages -= nr;
0187 }
0188 
0189 /*
0190  * Delete a page from the page cache and free it. Caller has to make
0191  * sure the page is locked and that nobody else uses it - or that usage
0192  * is safe.  The caller must hold the mapping's tree_lock.
0193  */
0194 void __delete_from_page_cache(struct page *page, void *shadow)
0195 {
0196     struct address_space *mapping = page->mapping;
0197     int nr = hpage_nr_pages(page);
0198 
0199     trace_mm_filemap_delete_from_page_cache(page);
0200     /*
0201      * if we're uptodate, flush out into the cleancache, otherwise
0202      * invalidate any existing cleancache entries.  We can't leave
0203      * stale data around in the cleancache once our page is gone
0204      */
0205     if (PageUptodate(page) && PageMappedToDisk(page))
0206         cleancache_put_page(page);
0207     else
0208         cleancache_invalidate_page(mapping, page);
0209 
0210     VM_BUG_ON_PAGE(PageTail(page), page);
0211     VM_BUG_ON_PAGE(page_mapped(page), page);
0212     if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
0213         int mapcount;
0214 
0215         pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
0216              current->comm, page_to_pfn(page));
0217         dump_page(page, "still mapped when deleted");
0218         dump_stack();
0219         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
0220 
0221         mapcount = page_mapcount(page);
0222         if (mapping_exiting(mapping) &&
0223             page_count(page) >= mapcount + 2) {
0224             /*
0225              * All vmas have already been torn down, so it's
0226              * a good bet that actually the page is unmapped,
0227              * and we'd prefer not to leak it: if we're wrong,
0228              * some other bad page check should catch it later.
0229              */
0230             page_mapcount_reset(page);
0231             page_ref_sub(page, mapcount);
0232         }
0233     }
0234 
0235     page_cache_tree_delete(mapping, page, shadow);
0236 
0237     page->mapping = NULL;
0238     /* Leave page->index set: truncation lookup relies upon it */
0239 
0240     /* hugetlb pages do not participate in page cache accounting. */
0241     if (!PageHuge(page))
0242         __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
0243     if (PageSwapBacked(page)) {
0244         __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
0245         if (PageTransHuge(page))
0246             __dec_node_page_state(page, NR_SHMEM_THPS);
0247     } else {
0248         VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
0249     }
0250 
0251     /*
0252      * At this point page must be either written or cleaned by truncate.
0253      * Dirty page here signals a bug and loss of unwritten data.
0254      *
0255      * This fixes dirty accounting after removing the page entirely but
0256      * leaves PageDirty set: it has no effect for truncated page and
0257      * anyway will be cleared before returning page into buddy allocator.
0258      */
0259     if (WARN_ON_ONCE(PageDirty(page)))
0260         account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
0261 }
0262 
0263 /**
0264  * delete_from_page_cache - delete page from page cache
0265  * @page: the page which the kernel is trying to remove from page cache
0266  *
0267  * This must be called only on pages that have been verified to be in the page
0268  * cache and locked.  It will never put the page into the free list, the caller
0269  * has a reference on the page.
0270  */
0271 void delete_from_page_cache(struct page *page)
0272 {
0273     struct address_space *mapping = page_mapping(page);
0274     unsigned long flags;
0275     void (*freepage)(struct page *);
0276 
0277     BUG_ON(!PageLocked(page));
0278 
0279     freepage = mapping->a_ops->freepage;
0280 
0281     spin_lock_irqsave(&mapping->tree_lock, flags);
0282     __delete_from_page_cache(page, NULL);
0283     spin_unlock_irqrestore(&mapping->tree_lock, flags);
0284 
0285     if (freepage)
0286         freepage(page);
0287 
0288     if (PageTransHuge(page) && !PageHuge(page)) {
0289         page_ref_sub(page, HPAGE_PMD_NR);
0290         VM_BUG_ON_PAGE(page_count(page) <= 0, page);
0291     } else {
0292         put_page(page);
0293     }
0294 }
0295 EXPORT_SYMBOL(delete_from_page_cache);
0296 
0297 int filemap_check_errors(struct address_space *mapping)
0298 {
0299     int ret = 0;
0300     /* Check for outstanding write errors */
0301     if (test_bit(AS_ENOSPC, &mapping->flags) &&
0302         test_and_clear_bit(AS_ENOSPC, &mapping->flags))
0303         ret = -ENOSPC;
0304     if (test_bit(AS_EIO, &mapping->flags) &&
0305         test_and_clear_bit(AS_EIO, &mapping->flags))
0306         ret = -EIO;
0307     return ret;
0308 }
0309 EXPORT_SYMBOL(filemap_check_errors);
0310 
0311 /**
0312  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
0313  * @mapping:    address space structure to write
0314  * @start:  offset in bytes where the range starts
0315  * @end:    offset in bytes where the range ends (inclusive)
0316  * @sync_mode:  enable synchronous operation
0317  *
0318  * Start writeback against all of a mapping's dirty pages that lie
0319  * within the byte offsets <start, end> inclusive.
0320  *
0321  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
0322  * opposed to a regular memory cleansing writeback.  The difference between
0323  * these two operations is that if a dirty page/buffer is encountered, it must
0324  * be waited upon, and not just skipped over.
0325  */
0326 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
0327                 loff_t end, int sync_mode)
0328 {
0329     int ret;
0330     struct writeback_control wbc = {
0331         .sync_mode = sync_mode,
0332         .nr_to_write = LONG_MAX,
0333         .range_start = start,
0334         .range_end = end,
0335     };
0336 
0337     if (!mapping_cap_writeback_dirty(mapping))
0338         return 0;
0339 
0340     wbc_attach_fdatawrite_inode(&wbc, mapping->host);
0341     ret = do_writepages(mapping, &wbc);
0342     wbc_detach_inode(&wbc);
0343     return ret;
0344 }
0345 
0346 static inline int __filemap_fdatawrite(struct address_space *mapping,
0347     int sync_mode)
0348 {
0349     return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
0350 }
0351 
0352 int filemap_fdatawrite(struct address_space *mapping)
0353 {
0354     return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
0355 }
0356 EXPORT_SYMBOL(filemap_fdatawrite);
0357 
0358 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
0359                 loff_t end)
0360 {
0361     return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
0362 }
0363 EXPORT_SYMBOL(filemap_fdatawrite_range);
0364 
0365 /**
0366  * filemap_flush - mostly a non-blocking flush
0367  * @mapping:    target address_space
0368  *
0369  * This is a mostly non-blocking flush.  Not suitable for data-integrity
0370  * purposes - I/O may not be started against all dirty pages.
0371  */
0372 int filemap_flush(struct address_space *mapping)
0373 {
0374     return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
0375 }
0376 EXPORT_SYMBOL(filemap_flush);
0377 
0378 static int __filemap_fdatawait_range(struct address_space *mapping,
0379                      loff_t start_byte, loff_t end_byte)
0380 {
0381     pgoff_t index = start_byte >> PAGE_SHIFT;
0382     pgoff_t end = end_byte >> PAGE_SHIFT;
0383     struct pagevec pvec;
0384     int nr_pages;
0385     int ret = 0;
0386 
0387     if (end_byte < start_byte)
0388         goto out;
0389 
0390     pagevec_init(&pvec, 0);
0391     while ((index <= end) &&
0392             (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
0393             PAGECACHE_TAG_WRITEBACK,
0394             min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
0395         unsigned i;
0396 
0397         for (i = 0; i < nr_pages; i++) {
0398             struct page *page = pvec.pages[i];
0399 
0400             /* until radix tree lookup accepts end_index */
0401             if (page->index > end)
0402                 continue;
0403 
0404             wait_on_page_writeback(page);
0405             if (TestClearPageError(page))
0406                 ret = -EIO;
0407         }
0408         pagevec_release(&pvec);
0409         cond_resched();
0410     }
0411 out:
0412     return ret;
0413 }
0414 
0415 /**
0416  * filemap_fdatawait_range - wait for writeback to complete
0417  * @mapping:        address space structure to wait for
0418  * @start_byte:     offset in bytes where the range starts
0419  * @end_byte:       offset in bytes where the range ends (inclusive)
0420  *
0421  * Walk the list of under-writeback pages of the given address space
0422  * in the given range and wait for all of them.  Check error status of
0423  * the address space and return it.
0424  *
0425  * Since the error status of the address space is cleared by this function,
0426  * callers are responsible for checking the return value and handling and/or
0427  * reporting the error.
0428  */
0429 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
0430                 loff_t end_byte)
0431 {
0432     int ret, ret2;
0433 
0434     ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
0435     ret2 = filemap_check_errors(mapping);
0436     if (!ret)
0437         ret = ret2;
0438 
0439     return ret;
0440 }
0441 EXPORT_SYMBOL(filemap_fdatawait_range);
0442 
0443 /**
0444  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
0445  * @mapping: address space structure to wait for
0446  *
0447  * Walk the list of under-writeback pages of the given address space
0448  * and wait for all of them.  Unlike filemap_fdatawait(), this function
0449  * does not clear error status of the address space.
0450  *
0451  * Use this function if callers don't handle errors themselves.  Expected
0452  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
0453  * fsfreeze(8)
0454  */
0455 void filemap_fdatawait_keep_errors(struct address_space *mapping)
0456 {
0457     loff_t i_size = i_size_read(mapping->host);
0458 
0459     if (i_size == 0)
0460         return;
0461 
0462     __filemap_fdatawait_range(mapping, 0, i_size - 1);
0463 }
0464 
0465 /**
0466  * filemap_fdatawait - wait for all under-writeback pages to complete
0467  * @mapping: address space structure to wait for
0468  *
0469  * Walk the list of under-writeback pages of the given address space
0470  * and wait for all of them.  Check error status of the address space
0471  * and return it.
0472  *
0473  * Since the error status of the address space is cleared by this function,
0474  * callers are responsible for checking the return value and handling and/or
0475  * reporting the error.
0476  */
0477 int filemap_fdatawait(struct address_space *mapping)
0478 {
0479     loff_t i_size = i_size_read(mapping->host);
0480 
0481     if (i_size == 0)
0482         return 0;
0483 
0484     return filemap_fdatawait_range(mapping, 0, i_size - 1);
0485 }
0486 EXPORT_SYMBOL(filemap_fdatawait);
0487 
0488 int filemap_write_and_wait(struct address_space *mapping)
0489 {
0490     int err = 0;
0491 
0492     if ((!dax_mapping(mapping) && mapping->nrpages) ||
0493         (dax_mapping(mapping) && mapping->nrexceptional)) {
0494         err = filemap_fdatawrite(mapping);
0495         /*
0496          * Even if the above returned error, the pages may be
0497          * written partially (e.g. -ENOSPC), so we wait for it.
0498          * But the -EIO is special case, it may indicate the worst
0499          * thing (e.g. bug) happened, so we avoid waiting for it.
0500          */
0501         if (err != -EIO) {
0502             int err2 = filemap_fdatawait(mapping);
0503             if (!err)
0504                 err = err2;
0505         }
0506     } else {
0507         err = filemap_check_errors(mapping);
0508     }
0509     return err;
0510 }
0511 EXPORT_SYMBOL(filemap_write_and_wait);
0512 
0513 /**
0514  * filemap_write_and_wait_range - write out & wait on a file range
0515  * @mapping:    the address_space for the pages
0516  * @lstart: offset in bytes where the range starts
0517  * @lend:   offset in bytes where the range ends (inclusive)
0518  *
0519  * Write out and wait upon file offsets lstart->lend, inclusive.
0520  *
0521  * Note that `lend' is inclusive (describes the last byte to be written) so
0522  * that this function can be used to write to the very end-of-file (end = -1).
0523  */
0524 int filemap_write_and_wait_range(struct address_space *mapping,
0525                  loff_t lstart, loff_t lend)
0526 {
0527     int err = 0;
0528 
0529     if ((!dax_mapping(mapping) && mapping->nrpages) ||
0530         (dax_mapping(mapping) && mapping->nrexceptional)) {
0531         err = __filemap_fdatawrite_range(mapping, lstart, lend,
0532                          WB_SYNC_ALL);
0533         /* See comment of filemap_write_and_wait() */
0534         if (err != -EIO) {
0535             int err2 = filemap_fdatawait_range(mapping,
0536                         lstart, lend);
0537             if (!err)
0538                 err = err2;
0539         }
0540     } else {
0541         err = filemap_check_errors(mapping);
0542     }
0543     return err;
0544 }
0545 EXPORT_SYMBOL(filemap_write_and_wait_range);
0546 
0547 /**
0548  * replace_page_cache_page - replace a pagecache page with a new one
0549  * @old:    page to be replaced
0550  * @new:    page to replace with
0551  * @gfp_mask:   allocation mode
0552  *
0553  * This function replaces a page in the pagecache with a new one.  On
0554  * success it acquires the pagecache reference for the new page and
0555  * drops it for the old page.  Both the old and new pages must be
0556  * locked.  This function does not add the new page to the LRU, the
0557  * caller must do that.
0558  *
0559  * The remove + add is atomic.  The only way this function can fail is
0560  * memory allocation failure.
0561  */
0562 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
0563 {
0564     int error;
0565 
0566     VM_BUG_ON_PAGE(!PageLocked(old), old);
0567     VM_BUG_ON_PAGE(!PageLocked(new), new);
0568     VM_BUG_ON_PAGE(new->mapping, new);
0569 
0570     error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
0571     if (!error) {
0572         struct address_space *mapping = old->mapping;
0573         void (*freepage)(struct page *);
0574         unsigned long flags;
0575 
0576         pgoff_t offset = old->index;
0577         freepage = mapping->a_ops->freepage;
0578 
0579         get_page(new);
0580         new->mapping = mapping;
0581         new->index = offset;
0582 
0583         spin_lock_irqsave(&mapping->tree_lock, flags);
0584         __delete_from_page_cache(old, NULL);
0585         error = page_cache_tree_insert(mapping, new, NULL);
0586         BUG_ON(error);
0587 
0588         /*
0589          * hugetlb pages do not participate in page cache accounting.
0590          */
0591         if (!PageHuge(new))
0592             __inc_node_page_state(new, NR_FILE_PAGES);
0593         if (PageSwapBacked(new))
0594             __inc_node_page_state(new, NR_SHMEM);
0595         spin_unlock_irqrestore(&mapping->tree_lock, flags);
0596         mem_cgroup_migrate(old, new);
0597         radix_tree_preload_end();
0598         if (freepage)
0599             freepage(old);
0600         put_page(old);
0601     }
0602 
0603     return error;
0604 }
0605 EXPORT_SYMBOL_GPL(replace_page_cache_page);
0606 
0607 static int __add_to_page_cache_locked(struct page *page,
0608                       struct address_space *mapping,
0609                       pgoff_t offset, gfp_t gfp_mask,
0610                       void **shadowp)
0611 {
0612     int huge = PageHuge(page);
0613     struct mem_cgroup *memcg;
0614     int error;
0615 
0616     VM_BUG_ON_PAGE(!PageLocked(page), page);
0617     VM_BUG_ON_PAGE(PageSwapBacked(page), page);
0618 
0619     if (!huge) {
0620         error = mem_cgroup_try_charge(page, current->mm,
0621                           gfp_mask, &memcg, false);
0622         if (error)
0623             return error;
0624     }
0625 
0626     error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
0627     if (error) {
0628         if (!huge)
0629             mem_cgroup_cancel_charge(page, memcg, false);
0630         return error;
0631     }
0632 
0633     get_page(page);
0634     page->mapping = mapping;
0635     page->index = offset;
0636 
0637     spin_lock_irq(&mapping->tree_lock);
0638     error = page_cache_tree_insert(mapping, page, shadowp);
0639     radix_tree_preload_end();
0640     if (unlikely(error))
0641         goto err_insert;
0642 
0643     /* hugetlb pages do not participate in page cache accounting. */
0644     if (!huge)
0645         __inc_node_page_state(page, NR_FILE_PAGES);
0646     spin_unlock_irq(&mapping->tree_lock);
0647     if (!huge)
0648         mem_cgroup_commit_charge(page, memcg, false, false);
0649     trace_mm_filemap_add_to_page_cache(page);
0650     return 0;
0651 err_insert:
0652     page->mapping = NULL;
0653     /* Leave page->index set: truncation relies upon it */
0654     spin_unlock_irq(&mapping->tree_lock);
0655     if (!huge)
0656         mem_cgroup_cancel_charge(page, memcg, false);
0657     put_page(page);
0658     return error;
0659 }
0660 
0661 /**
0662  * add_to_page_cache_locked - add a locked page to the pagecache
0663  * @page:   page to add
0664  * @mapping:    the page's address_space
0665  * @offset: page index
0666  * @gfp_mask:   page allocation mode
0667  *
0668  * This function is used to add a page to the pagecache. It must be locked.
0669  * This function does not add the page to the LRU.  The caller must do that.
0670  */
0671 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
0672         pgoff_t offset, gfp_t gfp_mask)
0673 {
0674     return __add_to_page_cache_locked(page, mapping, offset,
0675                       gfp_mask, NULL);
0676 }
0677 EXPORT_SYMBOL(add_to_page_cache_locked);
0678 
0679 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
0680                 pgoff_t offset, gfp_t gfp_mask)
0681 {
0682     void *shadow = NULL;
0683     int ret;
0684 
0685     __SetPageLocked(page);
0686     ret = __add_to_page_cache_locked(page, mapping, offset,
0687                      gfp_mask, &shadow);
0688     if (unlikely(ret))
0689         __ClearPageLocked(page);
0690     else {
0691         /*
0692          * The page might have been evicted from cache only
0693          * recently, in which case it should be activated like
0694          * any other repeatedly accessed page.
0695          * The exception is pages getting rewritten; evicting other
0696          * data from the working set, only to cache data that will
0697          * get overwritten with something else, is a waste of memory.
0698          */
0699         if (!(gfp_mask & __GFP_WRITE) &&
0700             shadow && workingset_refault(shadow)) {
0701             SetPageActive(page);
0702             workingset_activation(page);
0703         } else
0704             ClearPageActive(page);
0705         lru_cache_add(page);
0706     }
0707     return ret;
0708 }
0709 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
0710 
0711 #ifdef CONFIG_NUMA
0712 struct page *__page_cache_alloc(gfp_t gfp)
0713 {
0714     int n;
0715     struct page *page;
0716 
0717     if (cpuset_do_page_mem_spread()) {
0718         unsigned int cpuset_mems_cookie;
0719         do {
0720             cpuset_mems_cookie = read_mems_allowed_begin();
0721             n = cpuset_mem_spread_node();
0722             page = __alloc_pages_node(n, gfp, 0);
0723         } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
0724 
0725         return page;
0726     }
0727     return alloc_pages(gfp, 0);
0728 }
0729 EXPORT_SYMBOL(__page_cache_alloc);
0730 #endif
0731 
0732 /*
0733  * In order to wait for pages to become available there must be
0734  * waitqueues associated with pages. By using a hash table of
0735  * waitqueues where the bucket discipline is to maintain all
0736  * waiters on the same queue and wake all when any of the pages
0737  * become available, and for the woken contexts to check to be
0738  * sure the appropriate page became available, this saves space
0739  * at a cost of "thundering herd" phenomena during rare hash
0740  * collisions.
0741  */
0742 #define PAGE_WAIT_TABLE_BITS 8
0743 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
0744 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
0745 
0746 static wait_queue_head_t *page_waitqueue(struct page *page)
0747 {
0748     return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
0749 }
0750 
0751 void __init pagecache_init(void)
0752 {
0753     int i;
0754 
0755     for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
0756         init_waitqueue_head(&page_wait_table[i]);
0757 
0758     page_writeback_init();
0759 }
0760 
0761 struct wait_page_key {
0762     struct page *page;
0763     int bit_nr;
0764     int page_match;
0765 };
0766 
0767 struct wait_page_queue {
0768     struct page *page;
0769     int bit_nr;
0770     wait_queue_t wait;
0771 };
0772 
0773 static int wake_page_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
0774 {
0775     struct wait_page_key *key = arg;
0776     struct wait_page_queue *wait_page
0777         = container_of(wait, struct wait_page_queue, wait);
0778 
0779     if (wait_page->page != key->page)
0780            return 0;
0781     key->page_match = 1;
0782 
0783     if (wait_page->bit_nr != key->bit_nr)
0784         return 0;
0785     if (test_bit(key->bit_nr, &key->page->flags))
0786         return 0;
0787 
0788     return autoremove_wake_function(wait, mode, sync, key);
0789 }
0790 
0791 void wake_up_page_bit(struct page *page, int bit_nr)
0792 {
0793     wait_queue_head_t *q = page_waitqueue(page);
0794     struct wait_page_key key;
0795     unsigned long flags;
0796 
0797     key.page = page;
0798     key.bit_nr = bit_nr;
0799     key.page_match = 0;
0800 
0801     spin_lock_irqsave(&q->lock, flags);
0802     __wake_up_locked_key(q, TASK_NORMAL, &key);
0803     /*
0804      * It is possible for other pages to have collided on the waitqueue
0805      * hash, so in that case check for a page match. That prevents a long-
0806      * term waiter
0807      *
0808      * It is still possible to miss a case here, when we woke page waiters
0809      * and removed them from the waitqueue, but there are still other
0810      * page waiters.
0811      */
0812     if (!waitqueue_active(q) || !key.page_match) {
0813         ClearPageWaiters(page);
0814         /*
0815          * It's possible to miss clearing Waiters here, when we woke
0816          * our page waiters, but the hashed waitqueue has waiters for
0817          * other pages on it.
0818          *
0819          * That's okay, it's a rare case. The next waker will clear it.
0820          */
0821     }
0822     spin_unlock_irqrestore(&q->lock, flags);
0823 }
0824 EXPORT_SYMBOL(wake_up_page_bit);
0825 
0826 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
0827         struct page *page, int bit_nr, int state, bool lock)
0828 {
0829     struct wait_page_queue wait_page;
0830     wait_queue_t *wait = &wait_page.wait;
0831     int ret = 0;
0832 
0833     init_wait(wait);
0834     wait->func = wake_page_function;
0835     wait_page.page = page;
0836     wait_page.bit_nr = bit_nr;
0837 
0838     for (;;) {
0839         spin_lock_irq(&q->lock);
0840 
0841         if (likely(list_empty(&wait->task_list))) {
0842             if (lock)
0843                 __add_wait_queue_tail_exclusive(q, wait);
0844             else
0845                 __add_wait_queue(q, wait);
0846             SetPageWaiters(page);
0847         }
0848 
0849         set_current_state(state);
0850 
0851         spin_unlock_irq(&q->lock);
0852 
0853         if (likely(test_bit(bit_nr, &page->flags))) {
0854             io_schedule();
0855             if (unlikely(signal_pending_state(state, current))) {
0856                 ret = -EINTR;
0857                 break;
0858             }
0859         }
0860 
0861         if (lock) {
0862             if (!test_and_set_bit_lock(bit_nr, &page->flags))
0863                 break;
0864         } else {
0865             if (!test_bit(bit_nr, &page->flags))
0866                 break;
0867         }
0868     }
0869 
0870     finish_wait(q, wait);
0871 
0872     /*
0873      * A signal could leave PageWaiters set. Clearing it here if
0874      * !waitqueue_active would be possible (by open-coding finish_wait),
0875      * but still fail to catch it in the case of wait hash collision. We
0876      * already can fail to clear wait hash collision cases, so don't
0877      * bother with signals either.
0878      */
0879 
0880     return ret;
0881 }
0882 
0883 void wait_on_page_bit(struct page *page, int bit_nr)
0884 {
0885     wait_queue_head_t *q = page_waitqueue(page);
0886     wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
0887 }
0888 EXPORT_SYMBOL(wait_on_page_bit);
0889 
0890 int wait_on_page_bit_killable(struct page *page, int bit_nr)
0891 {
0892     wait_queue_head_t *q = page_waitqueue(page);
0893     return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
0894 }
0895 
0896 /**
0897  * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
0898  * @page: Page defining the wait queue of interest
0899  * @waiter: Waiter to add to the queue
0900  *
0901  * Add an arbitrary @waiter to the wait queue for the nominated @page.
0902  */
0903 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
0904 {
0905     wait_queue_head_t *q = page_waitqueue(page);
0906     unsigned long flags;
0907 
0908     spin_lock_irqsave(&q->lock, flags);
0909     __add_wait_queue(q, waiter);
0910     SetPageWaiters(page);
0911     spin_unlock_irqrestore(&q->lock, flags);
0912 }
0913 EXPORT_SYMBOL_GPL(add_page_wait_queue);
0914 
0915 #ifndef clear_bit_unlock_is_negative_byte
0916 
0917 /*
0918  * PG_waiters is the high bit in the same byte as PG_lock.
0919  *
0920  * On x86 (and on many other architectures), we can clear PG_lock and
0921  * test the sign bit at the same time. But if the architecture does
0922  * not support that special operation, we just do this all by hand
0923  * instead.
0924  *
0925  * The read of PG_waiters has to be after (or concurrently with) PG_locked
0926  * being cleared, but a memory barrier should be unneccssary since it is
0927  * in the same byte as PG_locked.
0928  */
0929 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
0930 {
0931     clear_bit_unlock(nr, mem);
0932     /* smp_mb__after_atomic(); */
0933     return test_bit(PG_waiters, mem);
0934 }
0935 
0936 #endif
0937 
0938 /**
0939  * unlock_page - unlock a locked page
0940  * @page: the page
0941  *
0942  * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
0943  * Also wakes sleepers in wait_on_page_writeback() because the wakeup
0944  * mechanism between PageLocked pages and PageWriteback pages is shared.
0945  * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
0946  *
0947  * Note that this depends on PG_waiters being the sign bit in the byte
0948  * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
0949  * clear the PG_locked bit and test PG_waiters at the same time fairly
0950  * portably (architectures that do LL/SC can test any bit, while x86 can
0951  * test the sign bit).
0952  */
0953 void unlock_page(struct page *page)
0954 {
0955     BUILD_BUG_ON(PG_waiters != 7);
0956     page = compound_head(page);
0957     VM_BUG_ON_PAGE(!PageLocked(page), page);
0958     if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
0959         wake_up_page_bit(page, PG_locked);
0960 }
0961 EXPORT_SYMBOL(unlock_page);
0962 
0963 /**
0964  * end_page_writeback - end writeback against a page
0965  * @page: the page
0966  */
0967 void end_page_writeback(struct page *page)
0968 {
0969     /*
0970      * TestClearPageReclaim could be used here but it is an atomic
0971      * operation and overkill in this particular case. Failing to
0972      * shuffle a page marked for immediate reclaim is too mild to
0973      * justify taking an atomic operation penalty at the end of
0974      * ever page writeback.
0975      */
0976     if (PageReclaim(page)) {
0977         ClearPageReclaim(page);
0978         rotate_reclaimable_page(page);
0979     }
0980 
0981     if (!test_clear_page_writeback(page))
0982         BUG();
0983 
0984     smp_mb__after_atomic();
0985     wake_up_page(page, PG_writeback);
0986 }
0987 EXPORT_SYMBOL(end_page_writeback);
0988 
0989 /*
0990  * After completing I/O on a page, call this routine to update the page
0991  * flags appropriately
0992  */
0993 void page_endio(struct page *page, bool is_write, int err)
0994 {
0995     if (!is_write) {
0996         if (!err) {
0997             SetPageUptodate(page);
0998         } else {
0999             ClearPageUptodate(page);
1000             SetPageError(page);
1001         }
1002         unlock_page(page);
1003     } else {
1004         if (err) {
1005             SetPageError(page);
1006             if (page->mapping)
1007                 mapping_set_error(page->mapping, err);
1008         }
1009         end_page_writeback(page);
1010     }
1011 }
1012 EXPORT_SYMBOL_GPL(page_endio);
1013 
1014 /**
1015  * __lock_page - get a lock on the page, assuming we need to sleep to get it
1016  * @page: the page to lock
1017  */
1018 void __lock_page(struct page *__page)
1019 {
1020     struct page *page = compound_head(__page);
1021     wait_queue_head_t *q = page_waitqueue(page);
1022     wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1023 }
1024 EXPORT_SYMBOL(__lock_page);
1025 
1026 int __lock_page_killable(struct page *__page)
1027 {
1028     struct page *page = compound_head(__page);
1029     wait_queue_head_t *q = page_waitqueue(page);
1030     return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1031 }
1032 EXPORT_SYMBOL_GPL(__lock_page_killable);
1033 
1034 /*
1035  * Return values:
1036  * 1 - page is locked; mmap_sem is still held.
1037  * 0 - page is not locked.
1038  *     mmap_sem has been released (up_read()), unless flags had both
1039  *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1040  *     which case mmap_sem is still held.
1041  *
1042  * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1043  * with the page locked and the mmap_sem unperturbed.
1044  */
1045 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1046              unsigned int flags)
1047 {
1048     if (flags & FAULT_FLAG_ALLOW_RETRY) {
1049         /*
1050          * CAUTION! In this case, mmap_sem is not released
1051          * even though return 0.
1052          */
1053         if (flags & FAULT_FLAG_RETRY_NOWAIT)
1054             return 0;
1055 
1056         up_read(&mm->mmap_sem);
1057         if (flags & FAULT_FLAG_KILLABLE)
1058             wait_on_page_locked_killable(page);
1059         else
1060             wait_on_page_locked(page);
1061         return 0;
1062     } else {
1063         if (flags & FAULT_FLAG_KILLABLE) {
1064             int ret;
1065 
1066             ret = __lock_page_killable(page);
1067             if (ret) {
1068                 up_read(&mm->mmap_sem);
1069                 return 0;
1070             }
1071         } else
1072             __lock_page(page);
1073         return 1;
1074     }
1075 }
1076 
1077 /**
1078  * page_cache_next_hole - find the next hole (not-present entry)
1079  * @mapping: mapping
1080  * @index: index
1081  * @max_scan: maximum range to search
1082  *
1083  * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1084  * lowest indexed hole.
1085  *
1086  * Returns: the index of the hole if found, otherwise returns an index
1087  * outside of the set specified (in which case 'return - index >=
1088  * max_scan' will be true). In rare cases of index wrap-around, 0 will
1089  * be returned.
1090  *
1091  * page_cache_next_hole may be called under rcu_read_lock. However,
1092  * like radix_tree_gang_lookup, this will not atomically search a
1093  * snapshot of the tree at a single point in time. For example, if a
1094  * hole is created at index 5, then subsequently a hole is created at
1095  * index 10, page_cache_next_hole covering both indexes may return 10
1096  * if called under rcu_read_lock.
1097  */
1098 pgoff_t page_cache_next_hole(struct address_space *mapping,
1099                  pgoff_t index, unsigned long max_scan)
1100 {
1101     unsigned long i;
1102 
1103     for (i = 0; i < max_scan; i++) {
1104         struct page *page;
1105 
1106         page = radix_tree_lookup(&mapping->page_tree, index);
1107         if (!page || radix_tree_exceptional_entry(page))
1108             break;
1109         index++;
1110         if (index == 0)
1111             break;
1112     }
1113 
1114     return index;
1115 }
1116 EXPORT_SYMBOL(page_cache_next_hole);
1117 
1118 /**
1119  * page_cache_prev_hole - find the prev hole (not-present entry)
1120  * @mapping: mapping
1121  * @index: index
1122  * @max_scan: maximum range to search
1123  *
1124  * Search backwards in the range [max(index-max_scan+1, 0), index] for
1125  * the first hole.
1126  *
1127  * Returns: the index of the hole if found, otherwise returns an index
1128  * outside of the set specified (in which case 'index - return >=
1129  * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1130  * will be returned.
1131  *
1132  * page_cache_prev_hole may be called under rcu_read_lock. However,
1133  * like radix_tree_gang_lookup, this will not atomically search a
1134  * snapshot of the tree at a single point in time. For example, if a
1135  * hole is created at index 10, then subsequently a hole is created at
1136  * index 5, page_cache_prev_hole covering both indexes may return 5 if
1137  * called under rcu_read_lock.
1138  */
1139 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1140                  pgoff_t index, unsigned long max_scan)
1141 {
1142     unsigned long i;
1143 
1144     for (i = 0; i < max_scan; i++) {
1145         struct page *page;
1146 
1147         page = radix_tree_lookup(&mapping->page_tree, index);
1148         if (!page || radix_tree_exceptional_entry(page))
1149             break;
1150         index--;
1151         if (index == ULONG_MAX)
1152             break;
1153     }
1154 
1155     return index;
1156 }
1157 EXPORT_SYMBOL(page_cache_prev_hole);
1158 
1159 /**
1160  * find_get_entry - find and get a page cache entry
1161  * @mapping: the address_space to search
1162  * @offset: the page cache index
1163  *
1164  * Looks up the page cache slot at @mapping & @offset.  If there is a
1165  * page cache page, it is returned with an increased refcount.
1166  *
1167  * If the slot holds a shadow entry of a previously evicted page, or a
1168  * swap entry from shmem/tmpfs, it is returned.
1169  *
1170  * Otherwise, %NULL is returned.
1171  */
1172 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1173 {
1174     void **pagep;
1175     struct page *head, *page;
1176 
1177     rcu_read_lock();
1178 repeat:
1179     page = NULL;
1180     pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1181     if (pagep) {
1182         page = radix_tree_deref_slot(pagep);
1183         if (unlikely(!page))
1184             goto out;
1185         if (radix_tree_exception(page)) {
1186             if (radix_tree_deref_retry(page))
1187                 goto repeat;
1188             /*
1189              * A shadow entry of a recently evicted page,
1190              * or a swap entry from shmem/tmpfs.  Return
1191              * it without attempting to raise page count.
1192              */
1193             goto out;
1194         }
1195 
1196         head = compound_head(page);
1197         if (!page_cache_get_speculative(head))
1198             goto repeat;
1199 
1200         /* The page was split under us? */
1201         if (compound_head(page) != head) {
1202             put_page(head);
1203             goto repeat;
1204         }
1205 
1206         /*
1207          * Has the page moved?
1208          * This is part of the lockless pagecache protocol. See
1209          * include/linux/pagemap.h for details.
1210          */
1211         if (unlikely(page != *pagep)) {
1212             put_page(head);
1213             goto repeat;
1214         }
1215     }
1216 out:
1217     rcu_read_unlock();
1218 
1219     return page;
1220 }
1221 EXPORT_SYMBOL(find_get_entry);
1222 
1223 /**
1224  * find_lock_entry - locate, pin and lock a page cache entry
1225  * @mapping: the address_space to search
1226  * @offset: the page cache index
1227  *
1228  * Looks up the page cache slot at @mapping & @offset.  If there is a
1229  * page cache page, it is returned locked and with an increased
1230  * refcount.
1231  *
1232  * If the slot holds a shadow entry of a previously evicted page, or a
1233  * swap entry from shmem/tmpfs, it is returned.
1234  *
1235  * Otherwise, %NULL is returned.
1236  *
1237  * find_lock_entry() may sleep.
1238  */
1239 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1240 {
1241     struct page *page;
1242 
1243 repeat:
1244     page = find_get_entry(mapping, offset);
1245     if (page && !radix_tree_exception(page)) {
1246         lock_page(page);
1247         /* Has the page been truncated? */
1248         if (unlikely(page_mapping(page) != mapping)) {
1249             unlock_page(page);
1250             put_page(page);
1251             goto repeat;
1252         }
1253         VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1254     }
1255     return page;
1256 }
1257 EXPORT_SYMBOL(find_lock_entry);
1258 
1259 /**
1260  * pagecache_get_page - find and get a page reference
1261  * @mapping: the address_space to search
1262  * @offset: the page index
1263  * @fgp_flags: PCG flags
1264  * @gfp_mask: gfp mask to use for the page cache data page allocation
1265  *
1266  * Looks up the page cache slot at @mapping & @offset.
1267  *
1268  * PCG flags modify how the page is returned.
1269  *
1270  * FGP_ACCESSED: the page will be marked accessed
1271  * FGP_LOCK: Page is return locked
1272  * FGP_CREAT: If page is not present then a new page is allocated using
1273  *      @gfp_mask and added to the page cache and the VM's LRU
1274  *      list. The page is returned locked and with an increased
1275  *      refcount. Otherwise, %NULL is returned.
1276  *
1277  * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1278  * if the GFP flags specified for FGP_CREAT are atomic.
1279  *
1280  * If there is a page cache page, it is returned with an increased refcount.
1281  */
1282 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1283     int fgp_flags, gfp_t gfp_mask)
1284 {
1285     struct page *page;
1286 
1287 repeat:
1288     page = find_get_entry(mapping, offset);
1289     if (radix_tree_exceptional_entry(page))
1290         page = NULL;
1291     if (!page)
1292         goto no_page;
1293 
1294     if (fgp_flags & FGP_LOCK) {
1295         if (fgp_flags & FGP_NOWAIT) {
1296             if (!trylock_page(page)) {
1297                 put_page(page);
1298                 return NULL;
1299             }
1300         } else {
1301             lock_page(page);
1302         }
1303 
1304         /* Has the page been truncated? */
1305         if (unlikely(page->mapping != mapping)) {
1306             unlock_page(page);
1307             put_page(page);
1308             goto repeat;
1309         }
1310         VM_BUG_ON_PAGE(page->index != offset, page);
1311     }
1312 
1313     if (page && (fgp_flags & FGP_ACCESSED))
1314         mark_page_accessed(page);
1315 
1316 no_page:
1317     if (!page && (fgp_flags & FGP_CREAT)) {
1318         int err;
1319         if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1320             gfp_mask |= __GFP_WRITE;
1321         if (fgp_flags & FGP_NOFS)
1322             gfp_mask &= ~__GFP_FS;
1323 
1324         page = __page_cache_alloc(gfp_mask);
1325         if (!page)
1326             return NULL;
1327 
1328         if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1329             fgp_flags |= FGP_LOCK;
1330 
1331         /* Init accessed so avoid atomic mark_page_accessed later */
1332         if (fgp_flags & FGP_ACCESSED)
1333             __SetPageReferenced(page);
1334 
1335         err = add_to_page_cache_lru(page, mapping, offset,
1336                 gfp_mask & GFP_RECLAIM_MASK);
1337         if (unlikely(err)) {
1338             put_page(page);
1339             page = NULL;
1340             if (err == -EEXIST)
1341                 goto repeat;
1342         }
1343     }
1344 
1345     return page;
1346 }
1347 EXPORT_SYMBOL(pagecache_get_page);
1348 
1349 /**
1350  * find_get_entries - gang pagecache lookup
1351  * @mapping:    The address_space to search
1352  * @start:  The starting page cache index
1353  * @nr_entries: The maximum number of entries
1354  * @entries:    Where the resulting entries are placed
1355  * @indices:    The cache indices corresponding to the entries in @entries
1356  *
1357  * find_get_entries() will search for and return a group of up to
1358  * @nr_entries entries in the mapping.  The entries are placed at
1359  * @entries.  find_get_entries() takes a reference against any actual
1360  * pages it returns.
1361  *
1362  * The search returns a group of mapping-contiguous page cache entries
1363  * with ascending indexes.  There may be holes in the indices due to
1364  * not-present pages.
1365  *
1366  * Any shadow entries of evicted pages, or swap entries from
1367  * shmem/tmpfs, are included in the returned array.
1368  *
1369  * find_get_entries() returns the number of pages and shadow entries
1370  * which were found.
1371  */
1372 unsigned find_get_entries(struct address_space *mapping,
1373               pgoff_t start, unsigned int nr_entries,
1374               struct page **entries, pgoff_t *indices)
1375 {
1376     void **slot;
1377     unsigned int ret = 0;
1378     struct radix_tree_iter iter;
1379 
1380     if (!nr_entries)
1381         return 0;
1382 
1383     rcu_read_lock();
1384     radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1385         struct page *head, *page;
1386 repeat:
1387         page = radix_tree_deref_slot(slot);
1388         if (unlikely(!page))
1389             continue;
1390         if (radix_tree_exception(page)) {
1391             if (radix_tree_deref_retry(page)) {
1392                 slot = radix_tree_iter_retry(&iter);
1393                 continue;
1394             }
1395             /*
1396              * A shadow entry of a recently evicted page, a swap
1397              * entry from shmem/tmpfs or a DAX entry.  Return it
1398              * without attempting to raise page count.
1399              */
1400             goto export;
1401         }
1402 
1403         head = compound_head(page);
1404         if (!page_cache_get_speculative(head))
1405             goto repeat;
1406 
1407         /* The page was split under us? */
1408         if (compound_head(page) != head) {
1409             put_page(head);
1410             goto repeat;
1411         }
1412 
1413         /* Has the page moved? */
1414         if (unlikely(page != *slot)) {
1415             put_page(head);
1416             goto repeat;
1417         }
1418 export:
1419         indices[ret] = iter.index;
1420         entries[ret] = page;
1421         if (++ret == nr_entries)
1422             break;
1423     }
1424     rcu_read_unlock();
1425     return ret;
1426 }
1427 
1428 /**
1429  * find_get_pages - gang pagecache lookup
1430  * @mapping:    The address_space to search
1431  * @start:  The starting page index
1432  * @nr_pages:   The maximum number of pages
1433  * @pages:  Where the resulting pages are placed
1434  *
1435  * find_get_pages() will search for and return a group of up to
1436  * @nr_pages pages in the mapping.  The pages are placed at @pages.
1437  * find_get_pages() takes a reference against the returned pages.
1438  *
1439  * The search returns a group of mapping-contiguous pages with ascending
1440  * indexes.  There may be holes in the indices due to not-present pages.
1441  *
1442  * find_get_pages() returns the number of pages which were found.
1443  */
1444 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1445                 unsigned int nr_pages, struct page **pages)
1446 {
1447     struct radix_tree_iter iter;
1448     void **slot;
1449     unsigned ret = 0;
1450 
1451     if (unlikely(!nr_pages))
1452         return 0;
1453 
1454     rcu_read_lock();
1455     radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1456         struct page *head, *page;
1457 repeat:
1458         page = radix_tree_deref_slot(slot);
1459         if (unlikely(!page))
1460             continue;
1461 
1462         if (radix_tree_exception(page)) {
1463             if (radix_tree_deref_retry(page)) {
1464                 slot = radix_tree_iter_retry(&iter);
1465                 continue;
1466             }
1467             /*
1468              * A shadow entry of a recently evicted page,
1469              * or a swap entry from shmem/tmpfs.  Skip
1470              * over it.
1471              */
1472             continue;
1473         }
1474 
1475         head = compound_head(page);
1476         if (!page_cache_get_speculative(head))
1477             goto repeat;
1478 
1479         /* The page was split under us? */
1480         if (compound_head(page) != head) {
1481             put_page(head);
1482             goto repeat;
1483         }
1484 
1485         /* Has the page moved? */
1486         if (unlikely(page != *slot)) {
1487             put_page(head);
1488             goto repeat;
1489         }
1490 
1491         pages[ret] = page;
1492         if (++ret == nr_pages)
1493             break;
1494     }
1495 
1496     rcu_read_unlock();
1497     return ret;
1498 }
1499 
1500 /**
1501  * find_get_pages_contig - gang contiguous pagecache lookup
1502  * @mapping:    The address_space to search
1503  * @index:  The starting page index
1504  * @nr_pages:   The maximum number of pages
1505  * @pages:  Where the resulting pages are placed
1506  *
1507  * find_get_pages_contig() works exactly like find_get_pages(), except
1508  * that the returned number of pages are guaranteed to be contiguous.
1509  *
1510  * find_get_pages_contig() returns the number of pages which were found.
1511  */
1512 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1513                    unsigned int nr_pages, struct page **pages)
1514 {
1515     struct radix_tree_iter iter;
1516     void **slot;
1517     unsigned int ret = 0;
1518 
1519     if (unlikely(!nr_pages))
1520         return 0;
1521 
1522     rcu_read_lock();
1523     radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1524         struct page *head, *page;
1525 repeat:
1526         page = radix_tree_deref_slot(slot);
1527         /* The hole, there no reason to continue */
1528         if (unlikely(!page))
1529             break;
1530 
1531         if (radix_tree_exception(page)) {
1532             if (radix_tree_deref_retry(page)) {
1533                 slot = radix_tree_iter_retry(&iter);
1534                 continue;
1535             }
1536             /*
1537              * A shadow entry of a recently evicted page,
1538              * or a swap entry from shmem/tmpfs.  Stop
1539              * looking for contiguous pages.
1540              */
1541             break;
1542         }
1543 
1544         head = compound_head(page);
1545         if (!page_cache_get_speculative(head))
1546             goto repeat;
1547 
1548         /* The page was split under us? */
1549         if (compound_head(page) != head) {
1550             put_page(head);
1551             goto repeat;
1552         }
1553 
1554         /* Has the page moved? */
1555         if (unlikely(page != *slot)) {
1556             put_page(head);
1557             goto repeat;
1558         }
1559 
1560         /*
1561          * must check mapping and index after taking the ref.
1562          * otherwise we can get both false positives and false
1563          * negatives, which is just confusing to the caller.
1564          */
1565         if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1566             put_page(page);
1567             break;
1568         }
1569 
1570         pages[ret] = page;
1571         if (++ret == nr_pages)
1572             break;
1573     }
1574     rcu_read_unlock();
1575     return ret;
1576 }
1577 EXPORT_SYMBOL(find_get_pages_contig);
1578 
1579 /**
1580  * find_get_pages_tag - find and return pages that match @tag
1581  * @mapping:    the address_space to search
1582  * @index:  the starting page index
1583  * @tag:    the tag index
1584  * @nr_pages:   the maximum number of pages
1585  * @pages:  where the resulting pages are placed
1586  *
1587  * Like find_get_pages, except we only return pages which are tagged with
1588  * @tag.   We update @index to index the next page for the traversal.
1589  */
1590 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1591             int tag, unsigned int nr_pages, struct page **pages)
1592 {
1593     struct radix_tree_iter iter;
1594     void **slot;
1595     unsigned ret = 0;
1596 
1597     if (unlikely(!nr_pages))
1598         return 0;
1599 
1600     rcu_read_lock();
1601     radix_tree_for_each_tagged(slot, &mapping->page_tree,
1602                    &iter, *index, tag) {
1603         struct page *head, *page;
1604 repeat:
1605         page = radix_tree_deref_slot(slot);
1606         if (unlikely(!page))
1607             continue;
1608 
1609         if (radix_tree_exception(page)) {
1610             if (radix_tree_deref_retry(page)) {
1611                 slot = radix_tree_iter_retry(&iter);
1612                 continue;
1613             }
1614             /*
1615              * A shadow entry of a recently evicted page.
1616              *
1617              * Those entries should never be tagged, but
1618              * this tree walk is lockless and the tags are
1619              * looked up in bulk, one radix tree node at a
1620              * time, so there is a sizable window for page
1621              * reclaim to evict a page we saw tagged.
1622              *
1623              * Skip over it.
1624              */
1625             continue;
1626         }
1627 
1628         head = compound_head(page);
1629         if (!page_cache_get_speculative(head))
1630             goto repeat;
1631 
1632         /* The page was split under us? */
1633         if (compound_head(page) != head) {
1634             put_page(head);
1635             goto repeat;
1636         }
1637 
1638         /* Has the page moved? */
1639         if (unlikely(page != *slot)) {
1640             put_page(head);
1641             goto repeat;
1642         }
1643 
1644         pages[ret] = page;
1645         if (++ret == nr_pages)
1646             break;
1647     }
1648 
1649     rcu_read_unlock();
1650 
1651     if (ret)
1652         *index = pages[ret - 1]->index + 1;
1653 
1654     return ret;
1655 }
1656 EXPORT_SYMBOL(find_get_pages_tag);
1657 
1658 /**
1659  * find_get_entries_tag - find and return entries that match @tag
1660  * @mapping:    the address_space to search
1661  * @start:  the starting page cache index
1662  * @tag:    the tag index
1663  * @nr_entries: the maximum number of entries
1664  * @entries:    where the resulting entries are placed
1665  * @indices:    the cache indices corresponding to the entries in @entries
1666  *
1667  * Like find_get_entries, except we only return entries which are tagged with
1668  * @tag.
1669  */
1670 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1671             int tag, unsigned int nr_entries,
1672             struct page **entries, pgoff_t *indices)
1673 {
1674     void **slot;
1675     unsigned int ret = 0;
1676     struct radix_tree_iter iter;
1677 
1678     if (!nr_entries)
1679         return 0;
1680 
1681     rcu_read_lock();
1682     radix_tree_for_each_tagged(slot, &mapping->page_tree,
1683                    &iter, start, tag) {
1684         struct page *head, *page;
1685 repeat:
1686         page = radix_tree_deref_slot(slot);
1687         if (unlikely(!page))
1688             continue;
1689         if (radix_tree_exception(page)) {
1690             if (radix_tree_deref_retry(page)) {
1691                 slot = radix_tree_iter_retry(&iter);
1692                 continue;
1693             }
1694 
1695             /*
1696              * A shadow entry of a recently evicted page, a swap
1697              * entry from shmem/tmpfs or a DAX entry.  Return it
1698              * without attempting to raise page count.
1699              */
1700             goto export;
1701         }
1702 
1703         head = compound_head(page);
1704         if (!page_cache_get_speculative(head))
1705             goto repeat;
1706 
1707         /* The page was split under us? */
1708         if (compound_head(page) != head) {
1709             put_page(head);
1710             goto repeat;
1711         }
1712 
1713         /* Has the page moved? */
1714         if (unlikely(page != *slot)) {
1715             put_page(head);
1716             goto repeat;
1717         }
1718 export:
1719         indices[ret] = iter.index;
1720         entries[ret] = page;
1721         if (++ret == nr_entries)
1722             break;
1723     }
1724     rcu_read_unlock();
1725     return ret;
1726 }
1727 EXPORT_SYMBOL(find_get_entries_tag);
1728 
1729 /*
1730  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1731  * a _large_ part of the i/o request. Imagine the worst scenario:
1732  *
1733  *      ---R__________________________________________B__________
1734  *         ^ reading here                             ^ bad block(assume 4k)
1735  *
1736  * read(R) => miss => readahead(R...B) => media error => frustrating retries
1737  * => failing the whole request => read(R) => read(R+1) =>
1738  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1739  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1740  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1741  *
1742  * It is going insane. Fix it by quickly scaling down the readahead size.
1743  */
1744 static void shrink_readahead_size_eio(struct file *filp,
1745                     struct file_ra_state *ra)
1746 {
1747     ra->ra_pages /= 4;
1748 }
1749 
1750 /**
1751  * do_generic_file_read - generic file read routine
1752  * @filp:   the file to read
1753  * @ppos:   current file position
1754  * @iter:   data destination
1755  * @written:    already copied
1756  *
1757  * This is a generic file read routine, and uses the
1758  * mapping->a_ops->readpage() function for the actual low-level stuff.
1759  *
1760  * This is really ugly. But the goto's actually try to clarify some
1761  * of the logic when it comes to error handling etc.
1762  */
1763 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1764         struct iov_iter *iter, ssize_t written)
1765 {
1766     struct address_space *mapping = filp->f_mapping;
1767     struct inode *inode = mapping->host;
1768     struct file_ra_state *ra = &filp->f_ra;
1769     pgoff_t index;
1770     pgoff_t last_index;
1771     pgoff_t prev_index;
1772     unsigned long offset;      /* offset into pagecache page */
1773     unsigned int prev_offset;
1774     int error = 0;
1775 
1776     if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1777         return 0;
1778     iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1779 
1780     index = *ppos >> PAGE_SHIFT;
1781     prev_index = ra->prev_pos >> PAGE_SHIFT;
1782     prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1783     last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1784     offset = *ppos & ~PAGE_MASK;
1785 
1786     for (;;) {
1787         struct page *page;
1788         pgoff_t end_index;
1789         loff_t isize;
1790         unsigned long nr, ret;
1791 
1792         cond_resched();
1793 find_page:
1794         if (fatal_signal_pending(current)) {
1795             error = -EINTR;
1796             goto out;
1797         }
1798 
1799         page = find_get_page(mapping, index);
1800         if (!page) {
1801             page_cache_sync_readahead(mapping,
1802                     ra, filp,
1803                     index, last_index - index);
1804             page = find_get_page(mapping, index);
1805             if (unlikely(page == NULL))
1806                 goto no_cached_page;
1807         }
1808         if (PageReadahead(page)) {
1809             page_cache_async_readahead(mapping,
1810                     ra, filp, page,
1811                     index, last_index - index);
1812         }
1813         if (!PageUptodate(page)) {
1814             /*
1815              * See comment in do_read_cache_page on why
1816              * wait_on_page_locked is used to avoid unnecessarily
1817              * serialisations and why it's safe.
1818              */
1819             error = wait_on_page_locked_killable(page);
1820             if (unlikely(error))
1821                 goto readpage_error;
1822             if (PageUptodate(page))
1823                 goto page_ok;
1824 
1825             if (inode->i_blkbits == PAGE_SHIFT ||
1826                     !mapping->a_ops->is_partially_uptodate)
1827                 goto page_not_up_to_date;
1828             /* pipes can't handle partially uptodate pages */
1829             if (unlikely(iter->type & ITER_PIPE))
1830                 goto page_not_up_to_date;
1831             if (!trylock_page(page))
1832                 goto page_not_up_to_date;
1833             /* Did it get truncated before we got the lock? */
1834             if (!page->mapping)
1835                 goto page_not_up_to_date_locked;
1836             if (!mapping->a_ops->is_partially_uptodate(page,
1837                             offset, iter->count))
1838                 goto page_not_up_to_date_locked;
1839             unlock_page(page);
1840         }
1841 page_ok:
1842         /*
1843          * i_size must be checked after we know the page is Uptodate.
1844          *
1845          * Checking i_size after the check allows us to calculate
1846          * the correct value for "nr", which means the zero-filled
1847          * part of the page is not copied back to userspace (unless
1848          * another truncate extends the file - this is desired though).
1849          */
1850 
1851         isize = i_size_read(inode);
1852         end_index = (isize - 1) >> PAGE_SHIFT;
1853         if (unlikely(!isize || index > end_index)) {
1854             put_page(page);
1855             goto out;
1856         }
1857 
1858         /* nr is the maximum number of bytes to copy from this page */
1859         nr = PAGE_SIZE;
1860         if (index == end_index) {
1861             nr = ((isize - 1) & ~PAGE_MASK) + 1;
1862             if (nr <= offset) {
1863                 put_page(page);
1864                 goto out;
1865             }
1866         }
1867         nr = nr - offset;
1868 
1869         /* If users can be writing to this page using arbitrary
1870          * virtual addresses, take care about potential aliasing
1871          * before reading the page on the kernel side.
1872          */
1873         if (mapping_writably_mapped(mapping))
1874             flush_dcache_page(page);
1875 
1876         /*
1877          * When a sequential read accesses a page several times,
1878          * only mark it as accessed the first time.
1879          */
1880         if (prev_index != index || offset != prev_offset)
1881             mark_page_accessed(page);
1882         prev_index = index;
1883 
1884         /*
1885          * Ok, we have the page, and it's up-to-date, so
1886          * now we can copy it to user space...
1887          */
1888 
1889         ret = copy_page_to_iter(page, offset, nr, iter);
1890         offset += ret;
1891         index += offset >> PAGE_SHIFT;
1892         offset &= ~PAGE_MASK;
1893         prev_offset = offset;
1894 
1895         put_page(page);
1896         written += ret;
1897         if (!iov_iter_count(iter))
1898             goto out;
1899         if (ret < nr) {
1900             error = -EFAULT;
1901             goto out;
1902         }
1903         continue;
1904 
1905 page_not_up_to_date:
1906         /* Get exclusive access to the page ... */
1907         error = lock_page_killable(page);
1908         if (unlikely(error))
1909             goto readpage_error;
1910 
1911 page_not_up_to_date_locked:
1912         /* Did it get truncated before we got the lock? */
1913         if (!page->mapping) {
1914             unlock_page(page);
1915             put_page(page);
1916             continue;
1917         }
1918 
1919         /* Did somebody else fill it already? */
1920         if (PageUptodate(page)) {
1921             unlock_page(page);
1922             goto page_ok;
1923         }
1924 
1925 readpage:
1926         /*
1927          * A previous I/O error may have been due to temporary
1928          * failures, eg. multipath errors.
1929          * PG_error will be set again if readpage fails.
1930          */
1931         ClearPageError(page);
1932         /* Start the actual read. The read will unlock the page. */
1933         error = mapping->a_ops->readpage(filp, page);
1934 
1935         if (unlikely(error)) {
1936             if (error == AOP_TRUNCATED_PAGE) {
1937                 put_page(page);
1938                 error = 0;
1939                 goto find_page;
1940             }
1941             goto readpage_error;
1942         }
1943 
1944         if (!PageUptodate(page)) {
1945             error = lock_page_killable(page);
1946             if (unlikely(error))
1947                 goto readpage_error;
1948             if (!PageUptodate(page)) {
1949                 if (page->mapping == NULL) {
1950                     /*
1951                      * invalidate_mapping_pages got it
1952                      */
1953                     unlock_page(page);
1954                     put_page(page);
1955                     goto find_page;
1956                 }
1957                 unlock_page(page);
1958                 shrink_readahead_size_eio(filp, ra);
1959                 error = -EIO;
1960                 goto readpage_error;
1961             }
1962             unlock_page(page);
1963         }
1964 
1965         goto page_ok;
1966 
1967 readpage_error:
1968         /* UHHUH! A synchronous read error occurred. Report it */
1969         put_page(page);
1970         goto out;
1971 
1972 no_cached_page:
1973         /*
1974          * Ok, it wasn't cached, so we need to create a new
1975          * page..
1976          */
1977         page = page_cache_alloc_cold(mapping);
1978         if (!page) {
1979             error = -ENOMEM;
1980             goto out;
1981         }
1982         error = add_to_page_cache_lru(page, mapping, index,
1983                 mapping_gfp_constraint(mapping, GFP_KERNEL));
1984         if (error) {
1985             put_page(page);
1986             if (error == -EEXIST) {
1987                 error = 0;
1988                 goto find_page;
1989             }
1990             goto out;
1991         }
1992         goto readpage;
1993     }
1994 
1995 out:
1996     ra->prev_pos = prev_index;
1997     ra->prev_pos <<= PAGE_SHIFT;
1998     ra->prev_pos |= prev_offset;
1999 
2000     *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2001     file_accessed(filp);
2002     return written ? written : error;
2003 }
2004 
2005 /**
2006  * generic_file_read_iter - generic filesystem read routine
2007  * @iocb:   kernel I/O control block
2008  * @iter:   destination for the data read
2009  *
2010  * This is the "read_iter()" routine for all filesystems
2011  * that can use the page cache directly.
2012  */
2013 ssize_t
2014 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2015 {
2016     struct file *file = iocb->ki_filp;
2017     ssize_t retval = 0;
2018     size_t count = iov_iter_count(iter);
2019 
2020     if (!count)
2021         goto out; /* skip atime */
2022 
2023     if (iocb->ki_flags & IOCB_DIRECT) {
2024         struct address_space *mapping = file->f_mapping;
2025         struct inode *inode = mapping->host;
2026         struct iov_iter data = *iter;
2027         loff_t size;
2028 
2029         size = i_size_read(inode);
2030         retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
2031                     iocb->ki_pos + count - 1);
2032         if (retval < 0)
2033             goto out;
2034 
2035         file_accessed(file);
2036 
2037         retval = mapping->a_ops->direct_IO(iocb, &data);
2038         if (retval >= 0) {
2039             iocb->ki_pos += retval;
2040             iov_iter_advance(iter, retval);
2041         }
2042 
2043         /*
2044          * Btrfs can have a short DIO read if we encounter
2045          * compressed extents, so if there was an error, or if
2046          * we've already read everything we wanted to, or if
2047          * there was a short read because we hit EOF, go ahead
2048          * and return.  Otherwise fallthrough to buffered io for
2049          * the rest of the read.  Buffered reads will not work for
2050          * DAX files, so don't bother trying.
2051          */
2052         if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
2053             IS_DAX(inode))
2054             goto out;
2055     }
2056 
2057     retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
2058 out:
2059     return retval;
2060 }
2061 EXPORT_SYMBOL(generic_file_read_iter);
2062 
2063 #ifdef CONFIG_MMU
2064 /**
2065  * page_cache_read - adds requested page to the page cache if not already there
2066  * @file:   file to read
2067  * @offset: page index
2068  * @gfp_mask:   memory allocation flags
2069  *
2070  * This adds the requested page to the page cache if it isn't already there,
2071  * and schedules an I/O to read in its contents from disk.
2072  */
2073 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2074 {
2075     struct address_space *mapping = file->f_mapping;
2076     struct page *page;
2077     int ret;
2078 
2079     do {
2080         page = __page_cache_alloc(gfp_mask|__GFP_COLD);
2081         if (!page)
2082             return -ENOMEM;
2083 
2084         ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
2085         if (ret == 0)
2086             ret = mapping->a_ops->readpage(file, page);
2087         else if (ret == -EEXIST)
2088             ret = 0; /* losing race to add is OK */
2089 
2090         put_page(page);
2091 
2092     } while (ret == AOP_TRUNCATED_PAGE);
2093 
2094     return ret;
2095 }
2096 
2097 #define MMAP_LOTSAMISS  (100)
2098 
2099 /*
2100  * Synchronous readahead happens when we don't even find
2101  * a page in the page cache at all.
2102  */
2103 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2104                    struct file_ra_state *ra,
2105                    struct file *file,
2106                    pgoff_t offset)
2107 {
2108     struct address_space *mapping = file->f_mapping;
2109 
2110     /* If we don't want any read-ahead, don't bother */
2111     if (vma->vm_flags & VM_RAND_READ)
2112         return;
2113     if (!ra->ra_pages)
2114         return;
2115 
2116     if (vma->vm_flags & VM_SEQ_READ) {
2117         page_cache_sync_readahead(mapping, ra, file, offset,
2118                       ra->ra_pages);
2119         return;
2120     }
2121 
2122     /* Avoid banging the cache line if not needed */
2123     if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2124         ra->mmap_miss++;
2125 
2126     /*
2127      * Do we miss much more than hit in this file? If so,
2128      * stop bothering with read-ahead. It will only hurt.
2129      */
2130     if (ra->mmap_miss > MMAP_LOTSAMISS)
2131         return;
2132 
2133     /*
2134      * mmap read-around
2135      */
2136     ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2137     ra->size = ra->ra_pages;
2138     ra->async_size = ra->ra_pages / 4;
2139     ra_submit(ra, mapping, file);
2140 }
2141 
2142 /*
2143  * Asynchronous readahead happens when we find the page and PG_readahead,
2144  * so we want to possibly extend the readahead further..
2145  */
2146 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2147                     struct file_ra_state *ra,
2148                     struct file *file,
2149                     struct page *page,
2150                     pgoff_t offset)
2151 {
2152     struct address_space *mapping = file->f_mapping;
2153 
2154     /* If we don't want any read-ahead, don't bother */
2155     if (vma->vm_flags & VM_RAND_READ)
2156         return;
2157     if (ra->mmap_miss > 0)
2158         ra->mmap_miss--;
2159     if (PageReadahead(page))
2160         page_cache_async_readahead(mapping, ra, file,
2161                        page, offset, ra->ra_pages);
2162 }
2163 
2164 /**
2165  * filemap_fault - read in file data for page fault handling
2166  * @vma:    vma in which the fault was taken
2167  * @vmf:    struct vm_fault containing details of the fault
2168  *
2169  * filemap_fault() is invoked via the vma operations vector for a
2170  * mapped memory region to read in file data during a page fault.
2171  *
2172  * The goto's are kind of ugly, but this streamlines the normal case of having
2173  * it in the page cache, and handles the special cases reasonably without
2174  * having a lot of duplicated code.
2175  *
2176  * vma->vm_mm->mmap_sem must be held on entry.
2177  *
2178  * If our return value has VM_FAULT_RETRY set, it's because
2179  * lock_page_or_retry() returned 0.
2180  * The mmap_sem has usually been released in this case.
2181  * See __lock_page_or_retry() for the exception.
2182  *
2183  * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2184  * has not been released.
2185  *
2186  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2187  */
2188 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2189 {
2190     int error;
2191     struct file *file = vma->vm_file;
2192     struct address_space *mapping = file->f_mapping;
2193     struct file_ra_state *ra = &file->f_ra;
2194     struct inode *inode = mapping->host;
2195     pgoff_t offset = vmf->pgoff;
2196     struct page *page;
2197     loff_t size;
2198     int ret = 0;
2199 
2200     size = round_up(i_size_read(inode), PAGE_SIZE);
2201     if (offset >= size >> PAGE_SHIFT)
2202         return VM_FAULT_SIGBUS;
2203 
2204     /*
2205      * Do we have something in the page cache already?
2206      */
2207     page = find_get_page(mapping, offset);
2208     if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2209         /*
2210          * We found the page, so try async readahead before
2211          * waiting for the lock.
2212          */
2213         do_async_mmap_readahead(vma, ra, file, page, offset);
2214     } else if (!page) {
2215         /* No page in the page cache at all */
2216         do_sync_mmap_readahead(vma, ra, file, offset);
2217         count_vm_event(PGMAJFAULT);
2218         mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
2219         ret = VM_FAULT_MAJOR;
2220 retry_find:
2221         page = find_get_page(mapping, offset);
2222         if (!page)
2223             goto no_cached_page;
2224     }
2225 
2226     if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2227         put_page(page);
2228         return ret | VM_FAULT_RETRY;
2229     }
2230 
2231     /* Did it get truncated? */
2232     if (unlikely(page->mapping != mapping)) {
2233         unlock_page(page);
2234         put_page(page);
2235         goto retry_find;
2236     }
2237     VM_BUG_ON_PAGE(page->index != offset, page);
2238 
2239     /*
2240      * We have a locked page in the page cache, now we need to check
2241      * that it's up-to-date. If not, it is going to be due to an error.
2242      */
2243     if (unlikely(!PageUptodate(page)))
2244         goto page_not_uptodate;
2245 
2246     /*
2247      * Found the page and have a reference on it.
2248      * We must recheck i_size under page lock.
2249      */
2250     size = round_up(i_size_read(inode), PAGE_SIZE);
2251     if (unlikely(offset >= size >> PAGE_SHIFT)) {
2252         unlock_page(page);
2253         put_page(page);
2254         return VM_FAULT_SIGBUS;
2255     }
2256 
2257     vmf->page = page;
2258     return ret | VM_FAULT_LOCKED;
2259 
2260 no_cached_page:
2261     /*
2262      * We're only likely to ever get here if MADV_RANDOM is in
2263      * effect.
2264      */
2265     error = page_cache_read(file, offset, vmf->gfp_mask);
2266 
2267     /*
2268      * The page we want has now been added to the page cache.
2269      * In the unlikely event that someone removed it in the
2270      * meantime, we'll just come back here and read it again.
2271      */
2272     if (error >= 0)
2273         goto retry_find;
2274 
2275     /*
2276      * An error return from page_cache_read can result if the
2277      * system is low on memory, or a problem occurs while trying
2278      * to schedule I/O.
2279      */
2280     if (error == -ENOMEM)
2281         return VM_FAULT_OOM;
2282     return VM_FAULT_SIGBUS;
2283 
2284 page_not_uptodate:
2285     /*
2286      * Umm, take care of errors if the page isn't up-to-date.
2287      * Try to re-read it _once_. We do this synchronously,
2288      * because there really aren't any performance issues here
2289      * and we need to check for errors.
2290      */
2291     ClearPageError(page);
2292     error = mapping->a_ops->readpage(file, page);
2293     if (!error) {
2294         wait_on_page_locked(page);
2295         if (!PageUptodate(page))
2296             error = -EIO;
2297     }
2298     put_page(page);
2299 
2300     if (!error || error == AOP_TRUNCATED_PAGE)
2301         goto retry_find;
2302 
2303     /* Things didn't work out. Return zero to tell the mm layer so. */
2304     shrink_readahead_size_eio(file, ra);
2305     return VM_FAULT_SIGBUS;
2306 }
2307 EXPORT_SYMBOL(filemap_fault);
2308 
2309 void filemap_map_pages(struct vm_fault *vmf,
2310         pgoff_t start_pgoff, pgoff_t end_pgoff)
2311 {
2312     struct radix_tree_iter iter;
2313     void **slot;
2314     struct file *file = vmf->vma->vm_file;
2315     struct address_space *mapping = file->f_mapping;
2316     pgoff_t last_pgoff = start_pgoff;
2317     loff_t size;
2318     struct page *head, *page;
2319 
2320     rcu_read_lock();
2321     radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2322             start_pgoff) {
2323         if (iter.index > end_pgoff)
2324             break;
2325 repeat:
2326         page = radix_tree_deref_slot(slot);
2327         if (unlikely(!page))
2328             goto next;
2329         if (radix_tree_exception(page)) {
2330             if (radix_tree_deref_retry(page)) {
2331                 slot = radix_tree_iter_retry(&iter);
2332                 continue;
2333             }
2334             goto next;
2335         }
2336 
2337         head = compound_head(page);
2338         if (!page_cache_get_speculative(head))
2339             goto repeat;
2340 
2341         /* The page was split under us? */
2342         if (compound_head(page) != head) {
2343             put_page(head);
2344             goto repeat;
2345         }
2346 
2347         /* Has the page moved? */
2348         if (unlikely(page != *slot)) {
2349             put_page(head);
2350             goto repeat;
2351         }
2352 
2353         if (!PageUptodate(page) ||
2354                 PageReadahead(page) ||
2355                 PageHWPoison(page))
2356             goto skip;
2357         if (!trylock_page(page))
2358             goto skip;
2359 
2360         if (page->mapping != mapping || !PageUptodate(page))
2361             goto unlock;
2362 
2363         size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2364         if (page->index >= size >> PAGE_SHIFT)
2365             goto unlock;
2366 
2367         if (file->f_ra.mmap_miss > 0)
2368             file->f_ra.mmap_miss--;
2369 
2370         vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2371         if (vmf->pte)
2372             vmf->pte += iter.index - last_pgoff;
2373         last_pgoff = iter.index;
2374         if (alloc_set_pte(vmf, NULL, page))
2375             goto unlock;
2376         unlock_page(page);
2377         goto next;
2378 unlock:
2379         unlock_page(page);
2380 skip:
2381         put_page(page);
2382 next:
2383         /* Huge page is mapped? No need to proceed. */
2384         if (pmd_trans_huge(*vmf->pmd))
2385             break;
2386         if (iter.index == end_pgoff)
2387             break;
2388     }
2389     rcu_read_unlock();
2390 }
2391 EXPORT_SYMBOL(filemap_map_pages);
2392 
2393 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2394 {
2395     struct page *page = vmf->page;
2396     struct inode *inode = file_inode(vma->vm_file);
2397     int ret = VM_FAULT_LOCKED;
2398 
2399     sb_start_pagefault(inode->i_sb);
2400     file_update_time(vma->vm_file);
2401     lock_page(page);
2402     if (page->mapping != inode->i_mapping) {
2403         unlock_page(page);
2404         ret = VM_FAULT_NOPAGE;
2405         goto out;
2406     }
2407     /*
2408      * We mark the page dirty already here so that when freeze is in
2409      * progress, we are guaranteed that writeback during freezing will
2410      * see the dirty page and writeprotect it again.
2411      */
2412     set_page_dirty(page);
2413     wait_for_stable_page(page);
2414 out:
2415     sb_end_pagefault(inode->i_sb);
2416     return ret;
2417 }
2418 EXPORT_SYMBOL(filemap_page_mkwrite);
2419 
2420 const struct vm_operations_struct generic_file_vm_ops = {
2421     .fault      = filemap_fault,
2422     .map_pages  = filemap_map_pages,
2423     .page_mkwrite   = filemap_page_mkwrite,
2424 };
2425 
2426 /* This is used for a general mmap of a disk file */
2427 
2428 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2429 {
2430     struct address_space *mapping = file->f_mapping;
2431 
2432     if (!mapping->a_ops->readpage)
2433         return -ENOEXEC;
2434     file_accessed(file);
2435     vma->vm_ops = &generic_file_vm_ops;
2436     return 0;
2437 }
2438 
2439 /*
2440  * This is for filesystems which do not implement ->writepage.
2441  */
2442 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2443 {
2444     if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2445         return -EINVAL;
2446     return generic_file_mmap(file, vma);
2447 }
2448 #else
2449 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2450 {
2451     return -ENOSYS;
2452 }
2453 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2454 {
2455     return -ENOSYS;
2456 }
2457 #endif /* CONFIG_MMU */
2458 
2459 EXPORT_SYMBOL(generic_file_mmap);
2460 EXPORT_SYMBOL(generic_file_readonly_mmap);
2461 
2462 static struct page *wait_on_page_read(struct page *page)
2463 {
2464     if (!IS_ERR(page)) {
2465         wait_on_page_locked(page);
2466         if (!PageUptodate(page)) {
2467             put_page(page);
2468             page = ERR_PTR(-EIO);
2469         }
2470     }
2471     return page;
2472 }
2473 
2474 static struct page *do_read_cache_page(struct address_space *mapping,
2475                 pgoff_t index,
2476                 int (*filler)(void *, struct page *),
2477                 void *data,
2478                 gfp_t gfp)
2479 {
2480     struct page *page;
2481     int err;
2482 repeat:
2483     page = find_get_page(mapping, index);
2484     if (!page) {
2485         page = __page_cache_alloc(gfp | __GFP_COLD);
2486         if (!page)
2487             return ERR_PTR(-ENOMEM);
2488         err = add_to_page_cache_lru(page, mapping, index, gfp);
2489         if (unlikely(err)) {
2490             put_page(page);
2491             if (err == -EEXIST)
2492                 goto repeat;
2493             /* Presumably ENOMEM for radix tree node */
2494             return ERR_PTR(err);
2495         }
2496 
2497 filler:
2498         err = filler(data, page);
2499         if (err < 0) {
2500             put_page(page);
2501             return ERR_PTR(err);
2502         }
2503 
2504         page = wait_on_page_read(page);
2505         if (IS_ERR(page))
2506             return page;
2507         goto out;
2508     }
2509     if (PageUptodate(page))
2510         goto out;
2511 
2512     /*
2513      * Page is not up to date and may be locked due one of the following
2514      * case a: Page is being filled and the page lock is held
2515      * case b: Read/write error clearing the page uptodate status
2516      * case c: Truncation in progress (page locked)
2517      * case d: Reclaim in progress
2518      *
2519      * Case a, the page will be up to date when the page is unlocked.
2520      *    There is no need to serialise on the page lock here as the page
2521      *    is pinned so the lock gives no additional protection. Even if the
2522      *    the page is truncated, the data is still valid if PageUptodate as
2523      *    it's a race vs truncate race.
2524      * Case b, the page will not be up to date
2525      * Case c, the page may be truncated but in itself, the data may still
2526      *    be valid after IO completes as it's a read vs truncate race. The
2527      *    operation must restart if the page is not uptodate on unlock but
2528      *    otherwise serialising on page lock to stabilise the mapping gives
2529      *    no additional guarantees to the caller as the page lock is
2530      *    released before return.
2531      * Case d, similar to truncation. If reclaim holds the page lock, it
2532      *    will be a race with remove_mapping that determines if the mapping
2533      *    is valid on unlock but otherwise the data is valid and there is
2534      *    no need to serialise with page lock.
2535      *
2536      * As the page lock gives no additional guarantee, we optimistically
2537      * wait on the page to be unlocked and check if it's up to date and
2538      * use the page if it is. Otherwise, the page lock is required to
2539      * distinguish between the different cases. The motivation is that we
2540      * avoid spurious serialisations and wakeups when multiple processes
2541      * wait on the same page for IO to complete.
2542      */
2543     wait_on_page_locked(page);
2544     if (PageUptodate(page))
2545         goto out;
2546 
2547     /* Distinguish between all the cases under the safety of the lock */
2548     lock_page(page);
2549 
2550     /* Case c or d, restart the operation */
2551     if (!page->mapping) {
2552         unlock_page(page);
2553         put_page(page);
2554         goto repeat;
2555     }
2556 
2557     /* Someone else locked and filled the page in a very small window */
2558     if (PageUptodate(page)) {
2559         unlock_page(page);
2560         goto out;
2561     }
2562     goto filler;
2563 
2564 out:
2565     mark_page_accessed(page);
2566     return page;
2567 }
2568 
2569 /**
2570  * read_cache_page - read into page cache, fill it if needed
2571  * @mapping:    the page's address_space
2572  * @index:  the page index
2573  * @filler: function to perform the read
2574  * @data:   first arg to filler(data, page) function, often left as NULL
2575  *
2576  * Read into the page cache. If a page already exists, and PageUptodate() is
2577  * not set, try to fill the page and wait for it to become unlocked.
2578  *
2579  * If the page does not get brought uptodate, return -EIO.
2580  */
2581 struct page *read_cache_page(struct address_space *mapping,
2582                 pgoff_t index,
2583                 int (*filler)(void *, struct page *),
2584                 void *data)
2585 {
2586     return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2587 }
2588 EXPORT_SYMBOL(read_cache_page);
2589 
2590 /**
2591  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2592  * @mapping:    the page's address_space
2593  * @index:  the page index
2594  * @gfp:    the page allocator flags to use if allocating
2595  *
2596  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2597  * any new page allocations done using the specified allocation flags.
2598  *
2599  * If the page does not get brought uptodate, return -EIO.
2600  */
2601 struct page *read_cache_page_gfp(struct address_space *mapping,
2602                 pgoff_t index,
2603                 gfp_t gfp)
2604 {
2605     filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2606 
2607     return do_read_cache_page(mapping, index, filler, NULL, gfp);
2608 }
2609 EXPORT_SYMBOL(read_cache_page_gfp);
2610 
2611 /*
2612  * Performs necessary checks before doing a write
2613  *
2614  * Can adjust writing position or amount of bytes to write.
2615  * Returns appropriate error code that caller should return or
2616  * zero in case that write should be allowed.
2617  */
2618 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2619 {
2620     struct file *file = iocb->ki_filp;
2621     struct inode *inode = file->f_mapping->host;
2622     unsigned long limit = rlimit(RLIMIT_FSIZE);
2623     loff_t pos;
2624 
2625     if (!iov_iter_count(from))
2626         return 0;
2627 
2628     /* FIXME: this is for backwards compatibility with 2.4 */
2629     if (iocb->ki_flags & IOCB_APPEND)
2630         iocb->ki_pos = i_size_read(inode);
2631 
2632     pos = iocb->ki_pos;
2633 
2634     if (limit != RLIM_INFINITY) {
2635         if (iocb->ki_pos >= limit) {
2636             send_sig(SIGXFSZ, current, 0);
2637             return -EFBIG;
2638         }
2639         iov_iter_truncate(from, limit - (unsigned long)pos);
2640     }
2641 
2642     /*
2643      * LFS rule
2644      */
2645     if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2646                 !(file->f_flags & O_LARGEFILE))) {
2647         if (pos >= MAX_NON_LFS)
2648             return -EFBIG;
2649         iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2650     }
2651 
2652     /*
2653      * Are we about to exceed the fs block limit ?
2654      *
2655      * If we have written data it becomes a short write.  If we have
2656      * exceeded without writing data we send a signal and return EFBIG.
2657      * Linus frestrict idea will clean these up nicely..
2658      */
2659     if (unlikely(pos >= inode->i_sb->s_maxbytes))
2660         return -EFBIG;
2661 
2662     iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2663     return iov_iter_count(from);
2664 }
2665 EXPORT_SYMBOL(generic_write_checks);
2666 
2667 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2668                 loff_t pos, unsigned len, unsigned flags,
2669                 struct page **pagep, void **fsdata)
2670 {
2671     const struct address_space_operations *aops = mapping->a_ops;
2672 
2673     return aops->write_begin(file, mapping, pos, len, flags,
2674                             pagep, fsdata);
2675 }
2676 EXPORT_SYMBOL(pagecache_write_begin);
2677 
2678 int pagecache_write_end(struct file *file, struct address_space *mapping,
2679                 loff_t pos, unsigned len, unsigned copied,
2680                 struct page *page, void *fsdata)
2681 {
2682     const struct address_space_operations *aops = mapping->a_ops;
2683 
2684     return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2685 }
2686 EXPORT_SYMBOL(pagecache_write_end);
2687 
2688 ssize_t
2689 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2690 {
2691     struct file *file = iocb->ki_filp;
2692     struct address_space *mapping = file->f_mapping;
2693     struct inode    *inode = mapping->host;
2694     loff_t      pos = iocb->ki_pos;
2695     ssize_t     written;
2696     size_t      write_len;
2697     pgoff_t     end;
2698     struct iov_iter data;
2699 
2700     write_len = iov_iter_count(from);
2701     end = (pos + write_len - 1) >> PAGE_SHIFT;
2702 
2703     written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2704     if (written)
2705         goto out;
2706 
2707     /*
2708      * After a write we want buffered reads to be sure to go to disk to get
2709      * the new data.  We invalidate clean cached page from the region we're
2710      * about to write.  We do this *before* the write so that we can return
2711      * without clobbering -EIOCBQUEUED from ->direct_IO().
2712      */
2713     if (mapping->nrpages) {
2714         written = invalidate_inode_pages2_range(mapping,
2715                     pos >> PAGE_SHIFT, end);
2716         /*
2717          * If a page can not be invalidated, return 0 to fall back
2718          * to buffered write.
2719          */
2720         if (written) {
2721             if (written == -EBUSY)
2722                 return 0;
2723             goto out;
2724         }
2725     }
2726 
2727     data = *from;
2728     written = mapping->a_ops->direct_IO(iocb, &data);
2729 
2730     /*
2731      * Finally, try again to invalidate clean pages which might have been
2732      * cached by non-direct readahead, or faulted in by get_user_pages()
2733      * if the source of the write was an mmap'ed region of the file
2734      * we're writing.  Either one is a pretty crazy thing to do,
2735      * so we don't support it 100%.  If this invalidation
2736      * fails, tough, the write still worked...
2737      */
2738     if (mapping->nrpages) {
2739         invalidate_inode_pages2_range(mapping,
2740                           pos >> PAGE_SHIFT, end);
2741     }
2742 
2743     if (written > 0) {
2744         pos += written;
2745         iov_iter_advance(from, written);
2746         if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2747             i_size_write(inode, pos);
2748             mark_inode_dirty(inode);
2749         }
2750         iocb->ki_pos = pos;
2751     }
2752 out:
2753     return written;
2754 }
2755 EXPORT_SYMBOL(generic_file_direct_write);
2756 
2757 /*
2758  * Find or create a page at the given pagecache position. Return the locked
2759  * page. This function is specifically for buffered writes.
2760  */
2761 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2762                     pgoff_t index, unsigned flags)
2763 {
2764     struct page *page;
2765     int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2766 
2767     if (flags & AOP_FLAG_NOFS)
2768         fgp_flags |= FGP_NOFS;
2769 
2770     page = pagecache_get_page(mapping, index, fgp_flags,
2771             mapping_gfp_mask(mapping));
2772     if (page)
2773         wait_for_stable_page(page);
2774 
2775     return page;
2776 }
2777 EXPORT_SYMBOL(grab_cache_page_write_begin);
2778 
2779 ssize_t generic_perform_write(struct file *file,
2780                 struct iov_iter *i, loff_t pos)
2781 {
2782     struct address_space *mapping = file->f_mapping;
2783     const struct address_space_operations *a_ops = mapping->a_ops;
2784     long status = 0;
2785     ssize_t written = 0;
2786     unsigned int flags = 0;
2787 
2788     /*
2789      * Copies from kernel address space cannot fail (NFSD is a big user).
2790      */
2791     if (!iter_is_iovec(i))
2792         flags |= AOP_FLAG_UNINTERRUPTIBLE;
2793 
2794     do {
2795         struct page *page;
2796         unsigned long offset;   /* Offset into pagecache page */
2797         unsigned long bytes;    /* Bytes to write to page */
2798         size_t copied;      /* Bytes copied from user */
2799         void *fsdata;
2800 
2801         offset = (pos & (PAGE_SIZE - 1));
2802         bytes = min_t(unsigned long, PAGE_SIZE - offset,
2803                         iov_iter_count(i));
2804 
2805 again:
2806         /*
2807          * Bring in the user page that we will copy from _first_.
2808          * Otherwise there's a nasty deadlock on copying from the
2809          * same page as we're writing to, without it being marked
2810          * up-to-date.
2811          *
2812          * Not only is this an optimisation, but it is also required
2813          * to check that the address is actually valid, when atomic
2814          * usercopies are used, below.
2815          */
2816         if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2817             status = -EFAULT;
2818             break;
2819         }
2820 
2821         if (fatal_signal_pending(current)) {
2822             status = -EINTR;
2823             break;
2824         }
2825 
2826         status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2827                         &page, &fsdata);
2828         if (unlikely(status < 0))
2829             break;
2830 
2831         if (mapping_writably_mapped(mapping))
2832             flush_dcache_page(page);
2833 
2834         copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2835         flush_dcache_page(page);
2836 
2837         status = a_ops->write_end(file, mapping, pos, bytes, copied,
2838                         page, fsdata);
2839         if (unlikely(status < 0))
2840             break;
2841         copied = status;
2842 
2843         cond_resched();
2844 
2845         iov_iter_advance(i, copied);
2846         if (unlikely(copied == 0)) {
2847             /*
2848              * If we were unable to copy any data at all, we must
2849              * fall back to a single segment length write.
2850              *
2851              * If we didn't fallback here, we could livelock
2852              * because not all segments in the iov can be copied at
2853              * once without a pagefault.
2854              */
2855             bytes = min_t(unsigned long, PAGE_SIZE - offset,
2856                         iov_iter_single_seg_count(i));
2857             goto again;
2858         }
2859         pos += copied;
2860         written += copied;
2861 
2862         balance_dirty_pages_ratelimited(mapping);
2863     } while (iov_iter_count(i));
2864 
2865     return written ? written : status;
2866 }
2867 EXPORT_SYMBOL(generic_perform_write);
2868 
2869 /**
2870  * __generic_file_write_iter - write data to a file
2871  * @iocb:   IO state structure (file, offset, etc.)
2872  * @from:   iov_iter with data to write
2873  *
2874  * This function does all the work needed for actually writing data to a
2875  * file. It does all basic checks, removes SUID from the file, updates
2876  * modification times and calls proper subroutines depending on whether we
2877  * do direct IO or a standard buffered write.
2878  *
2879  * It expects i_mutex to be grabbed unless we work on a block device or similar
2880  * object which does not need locking at all.
2881  *
2882  * This function does *not* take care of syncing data in case of O_SYNC write.
2883  * A caller has to handle it. This is mainly due to the fact that we want to
2884  * avoid syncing under i_mutex.
2885  */
2886 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2887 {
2888     struct file *file = iocb->ki_filp;
2889     struct address_space * mapping = file->f_mapping;
2890     struct inode    *inode = mapping->host;
2891     ssize_t     written = 0;
2892     ssize_t     err;
2893     ssize_t     status;
2894 
2895     /* We can write back this queue in page reclaim */
2896     current->backing_dev_info = inode_to_bdi(inode);
2897     err = file_remove_privs(file);
2898     if (err)
2899         goto out;
2900 
2901     err = file_update_time(file);
2902     if (err)
2903         goto out;
2904 
2905     if (iocb->ki_flags & IOCB_DIRECT) {
2906         loff_t pos, endbyte;
2907 
2908         written = generic_file_direct_write(iocb, from);
2909         /*
2910          * If the write stopped short of completing, fall back to
2911          * buffered writes.  Some filesystems do this for writes to
2912          * holes, for example.  For DAX files, a buffered write will
2913          * not succeed (even if it did, DAX does not handle dirty
2914          * page-cache pages correctly).
2915          */
2916         if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2917             goto out;
2918 
2919         status = generic_perform_write(file, from, pos = iocb->ki_pos);
2920         /*
2921          * If generic_perform_write() returned a synchronous error
2922          * then we want to return the number of bytes which were
2923          * direct-written, or the error code if that was zero.  Note
2924          * that this differs from normal direct-io semantics, which
2925          * will return -EFOO even if some bytes were written.
2926          */
2927         if (unlikely(status < 0)) {
2928             err = status;
2929             goto out;
2930         }
2931         /*
2932          * We need to ensure that the page cache pages are written to
2933          * disk and invalidated to preserve the expected O_DIRECT
2934          * semantics.
2935          */
2936         endbyte = pos + status - 1;
2937         err = filemap_write_and_wait_range(mapping, pos, endbyte);
2938         if (err == 0) {
2939             iocb->ki_pos = endbyte + 1;
2940             written += status;
2941             invalidate_mapping_pages(mapping,
2942                          pos >> PAGE_SHIFT,
2943                          endbyte >> PAGE_SHIFT);
2944         } else {
2945             /*
2946              * We don't know how much we wrote, so just return
2947              * the number of bytes which were direct-written
2948              */
2949         }
2950     } else {
2951         written = generic_perform_write(file, from, iocb->ki_pos);
2952         if (likely(written > 0))
2953             iocb->ki_pos += written;
2954     }
2955 out:
2956     current->backing_dev_info = NULL;
2957     return written ? written : err;
2958 }
2959 EXPORT_SYMBOL(__generic_file_write_iter);
2960 
2961 /**
2962  * generic_file_write_iter - write data to a file
2963  * @iocb:   IO state structure
2964  * @from:   iov_iter with data to write
2965  *
2966  * This is a wrapper around __generic_file_write_iter() to be used by most
2967  * filesystems. It takes care of syncing the file in case of O_SYNC file
2968  * and acquires i_mutex as needed.
2969  */
2970 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2971 {
2972     struct file *file = iocb->ki_filp;
2973     struct inode *inode = file->f_mapping->host;
2974     ssize_t ret;
2975 
2976     inode_lock(inode);
2977     ret = generic_write_checks(iocb, from);
2978     if (ret > 0)
2979         ret = __generic_file_write_iter(iocb, from);
2980     inode_unlock(inode);
2981 
2982     if (ret > 0)
2983         ret = generic_write_sync(iocb, ret);
2984     return ret;
2985 }
2986 EXPORT_SYMBOL(generic_file_write_iter);
2987 
2988 /**
2989  * try_to_release_page() - release old fs-specific metadata on a page
2990  *
2991  * @page: the page which the kernel is trying to free
2992  * @gfp_mask: memory allocation flags (and I/O mode)
2993  *
2994  * The address_space is to try to release any data against the page
2995  * (presumably at page->private).  If the release was successful, return `1'.
2996  * Otherwise return zero.
2997  *
2998  * This may also be called if PG_fscache is set on a page, indicating that the
2999  * page is known to the local caching routines.
3000  *
3001  * The @gfp_mask argument specifies whether I/O may be performed to release
3002  * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3003  *
3004  */
3005 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3006 {
3007     struct address_space * const mapping = page->mapping;
3008 
3009     BUG_ON(!PageLocked(page));
3010     if (PageWriteback(page))
3011         return 0;
3012 
3013     if (mapping && mapping->a_ops->releasepage)
3014         return mapping->a_ops->releasepage(page, gfp_mask);
3015     return try_to_free_buffers(page);
3016 }
3017 
3018 EXPORT_SYMBOL(try_to_release_page);