Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  linux/mm/page_io.c
0004  *
0005  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
0006  *
0007  *  Swap reorganised 29.12.95, 
0008  *  Asynchronous swapping added 30.12.95. Stephen Tweedie
0009  *  Removed race in async swapping. 14.4.1996. Bruno Haible
0010  *  Add swap of shared pages through the page cache. 20.2.1998. Stephen Tweedie
0011  *  Always use brw_page, life becomes simpler. 12 May 1998 Eric Biederman
0012  */
0013 
0014 #include <linux/mm.h>
0015 #include <linux/kernel_stat.h>
0016 #include <linux/gfp.h>
0017 #include <linux/pagemap.h>
0018 #include <linux/swap.h>
0019 #include <linux/bio.h>
0020 #include <linux/swapops.h>
0021 #include <linux/buffer_head.h>
0022 #include <linux/writeback.h>
0023 #include <linux/frontswap.h>
0024 #include <linux/blkdev.h>
0025 #include <linux/psi.h>
0026 #include <linux/uio.h>
0027 #include <linux/sched/task.h>
0028 #include <linux/delayacct.h>
0029 #include "swap.h"
0030 
0031 void end_swap_bio_write(struct bio *bio)
0032 {
0033     struct page *page = bio_first_page_all(bio);
0034 
0035     if (bio->bi_status) {
0036         SetPageError(page);
0037         /*
0038          * We failed to write the page out to swap-space.
0039          * Re-dirty the page in order to avoid it being reclaimed.
0040          * Also print a dire warning that things will go BAD (tm)
0041          * very quickly.
0042          *
0043          * Also clear PG_reclaim to avoid folio_rotate_reclaimable()
0044          */
0045         set_page_dirty(page);
0046         pr_alert_ratelimited("Write-error on swap-device (%u:%u:%llu)\n",
0047                      MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
0048                      (unsigned long long)bio->bi_iter.bi_sector);
0049         ClearPageReclaim(page);
0050     }
0051     end_page_writeback(page);
0052     bio_put(bio);
0053 }
0054 
0055 static void end_swap_bio_read(struct bio *bio)
0056 {
0057     struct page *page = bio_first_page_all(bio);
0058     struct task_struct *waiter = bio->bi_private;
0059 
0060     if (bio->bi_status) {
0061         SetPageError(page);
0062         ClearPageUptodate(page);
0063         pr_alert_ratelimited("Read-error on swap-device (%u:%u:%llu)\n",
0064                      MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
0065                      (unsigned long long)bio->bi_iter.bi_sector);
0066         goto out;
0067     }
0068 
0069     SetPageUptodate(page);
0070 out:
0071     unlock_page(page);
0072     WRITE_ONCE(bio->bi_private, NULL);
0073     bio_put(bio);
0074     if (waiter) {
0075         blk_wake_io_task(waiter);
0076         put_task_struct(waiter);
0077     }
0078 }
0079 
0080 int generic_swapfile_activate(struct swap_info_struct *sis,
0081                 struct file *swap_file,
0082                 sector_t *span)
0083 {
0084     struct address_space *mapping = swap_file->f_mapping;
0085     struct inode *inode = mapping->host;
0086     unsigned blocks_per_page;
0087     unsigned long page_no;
0088     unsigned blkbits;
0089     sector_t probe_block;
0090     sector_t last_block;
0091     sector_t lowest_block = -1;
0092     sector_t highest_block = 0;
0093     int nr_extents = 0;
0094     int ret;
0095 
0096     blkbits = inode->i_blkbits;
0097     blocks_per_page = PAGE_SIZE >> blkbits;
0098 
0099     /*
0100      * Map all the blocks into the extent tree.  This code doesn't try
0101      * to be very smart.
0102      */
0103     probe_block = 0;
0104     page_no = 0;
0105     last_block = i_size_read(inode) >> blkbits;
0106     while ((probe_block + blocks_per_page) <= last_block &&
0107             page_no < sis->max) {
0108         unsigned block_in_page;
0109         sector_t first_block;
0110 
0111         cond_resched();
0112 
0113         first_block = probe_block;
0114         ret = bmap(inode, &first_block);
0115         if (ret || !first_block)
0116             goto bad_bmap;
0117 
0118         /*
0119          * It must be PAGE_SIZE aligned on-disk
0120          */
0121         if (first_block & (blocks_per_page - 1)) {
0122             probe_block++;
0123             goto reprobe;
0124         }
0125 
0126         for (block_in_page = 1; block_in_page < blocks_per_page;
0127                     block_in_page++) {
0128             sector_t block;
0129 
0130             block = probe_block + block_in_page;
0131             ret = bmap(inode, &block);
0132             if (ret || !block)
0133                 goto bad_bmap;
0134 
0135             if (block != first_block + block_in_page) {
0136                 /* Discontiguity */
0137                 probe_block++;
0138                 goto reprobe;
0139             }
0140         }
0141 
0142         first_block >>= (PAGE_SHIFT - blkbits);
0143         if (page_no) {  /* exclude the header page */
0144             if (first_block < lowest_block)
0145                 lowest_block = first_block;
0146             if (first_block > highest_block)
0147                 highest_block = first_block;
0148         }
0149 
0150         /*
0151          * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
0152          */
0153         ret = add_swap_extent(sis, page_no, 1, first_block);
0154         if (ret < 0)
0155             goto out;
0156         nr_extents += ret;
0157         page_no++;
0158         probe_block += blocks_per_page;
0159 reprobe:
0160         continue;
0161     }
0162     ret = nr_extents;
0163     *span = 1 + highest_block - lowest_block;
0164     if (page_no == 0)
0165         page_no = 1;    /* force Empty message */
0166     sis->max = page_no;
0167     sis->pages = page_no - 1;
0168     sis->highest_bit = page_no - 1;
0169 out:
0170     return ret;
0171 bad_bmap:
0172     pr_err("swapon: swapfile has holes\n");
0173     ret = -EINVAL;
0174     goto out;
0175 }
0176 
0177 /*
0178  * We may have stale swap cache pages in memory: notice
0179  * them here and get rid of the unnecessary final write.
0180  */
0181 int swap_writepage(struct page *page, struct writeback_control *wbc)
0182 {
0183     int ret = 0;
0184 
0185     if (try_to_free_swap(page)) {
0186         unlock_page(page);
0187         goto out;
0188     }
0189     /*
0190      * Arch code may have to preserve more data than just the page
0191      * contents, e.g. memory tags.
0192      */
0193     ret = arch_prepare_to_swap(page);
0194     if (ret) {
0195         set_page_dirty(page);
0196         unlock_page(page);
0197         goto out;
0198     }
0199     if (frontswap_store(page) == 0) {
0200         set_page_writeback(page);
0201         unlock_page(page);
0202         end_page_writeback(page);
0203         goto out;
0204     }
0205     ret = __swap_writepage(page, wbc, end_swap_bio_write);
0206 out:
0207     return ret;
0208 }
0209 
0210 static inline void count_swpout_vm_event(struct page *page)
0211 {
0212 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
0213     if (unlikely(PageTransHuge(page)))
0214         count_vm_event(THP_SWPOUT);
0215 #endif
0216     count_vm_events(PSWPOUT, thp_nr_pages(page));
0217 }
0218 
0219 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
0220 static void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
0221 {
0222     struct cgroup_subsys_state *css;
0223     struct mem_cgroup *memcg;
0224 
0225     memcg = page_memcg(page);
0226     if (!memcg)
0227         return;
0228 
0229     rcu_read_lock();
0230     css = cgroup_e_css(memcg->css.cgroup, &io_cgrp_subsys);
0231     bio_associate_blkg_from_css(bio, css);
0232     rcu_read_unlock();
0233 }
0234 #else
0235 #define bio_associate_blkg_from_page(bio, page)     do { } while (0)
0236 #endif /* CONFIG_MEMCG && CONFIG_BLK_CGROUP */
0237 
0238 struct swap_iocb {
0239     struct kiocb        iocb;
0240     struct bio_vec      bvec[SWAP_CLUSTER_MAX];
0241     int         pages;
0242     int         len;
0243 };
0244 static mempool_t *sio_pool;
0245 
0246 int sio_pool_init(void)
0247 {
0248     if (!sio_pool) {
0249         mempool_t *pool = mempool_create_kmalloc_pool(
0250             SWAP_CLUSTER_MAX, sizeof(struct swap_iocb));
0251         if (cmpxchg(&sio_pool, NULL, pool))
0252             mempool_destroy(pool);
0253     }
0254     if (!sio_pool)
0255         return -ENOMEM;
0256     return 0;
0257 }
0258 
0259 static void sio_write_complete(struct kiocb *iocb, long ret)
0260 {
0261     struct swap_iocb *sio = container_of(iocb, struct swap_iocb, iocb);
0262     struct page *page = sio->bvec[0].bv_page;
0263     int p;
0264 
0265     if (ret != sio->len) {
0266         /*
0267          * In the case of swap-over-nfs, this can be a
0268          * temporary failure if the system has limited
0269          * memory for allocating transmit buffers.
0270          * Mark the page dirty and avoid
0271          * folio_rotate_reclaimable but rate-limit the
0272          * messages but do not flag PageError like
0273          * the normal direct-to-bio case as it could
0274          * be temporary.
0275          */
0276         pr_err_ratelimited("Write error %ld on dio swapfile (%llu)\n",
0277                    ret, page_file_offset(page));
0278         for (p = 0; p < sio->pages; p++) {
0279             page = sio->bvec[p].bv_page;
0280             set_page_dirty(page);
0281             ClearPageReclaim(page);
0282         }
0283     } else {
0284         for (p = 0; p < sio->pages; p++)
0285             count_swpout_vm_event(sio->bvec[p].bv_page);
0286     }
0287 
0288     for (p = 0; p < sio->pages; p++)
0289         end_page_writeback(sio->bvec[p].bv_page);
0290 
0291     mempool_free(sio, sio_pool);
0292 }
0293 
0294 static int swap_writepage_fs(struct page *page, struct writeback_control *wbc)
0295 {
0296     struct swap_iocb *sio = NULL;
0297     struct swap_info_struct *sis = page_swap_info(page);
0298     struct file *swap_file = sis->swap_file;
0299     loff_t pos = page_file_offset(page);
0300 
0301     set_page_writeback(page);
0302     unlock_page(page);
0303     if (wbc->swap_plug)
0304         sio = *wbc->swap_plug;
0305     if (sio) {
0306         if (sio->iocb.ki_filp != swap_file ||
0307             sio->iocb.ki_pos + sio->len != pos) {
0308             swap_write_unplug(sio);
0309             sio = NULL;
0310         }
0311     }
0312     if (!sio) {
0313         sio = mempool_alloc(sio_pool, GFP_NOIO);
0314         init_sync_kiocb(&sio->iocb, swap_file);
0315         sio->iocb.ki_complete = sio_write_complete;
0316         sio->iocb.ki_pos = pos;
0317         sio->pages = 0;
0318         sio->len = 0;
0319     }
0320     sio->bvec[sio->pages].bv_page = page;
0321     sio->bvec[sio->pages].bv_len = thp_size(page);
0322     sio->bvec[sio->pages].bv_offset = 0;
0323     sio->len += thp_size(page);
0324     sio->pages += 1;
0325     if (sio->pages == ARRAY_SIZE(sio->bvec) || !wbc->swap_plug) {
0326         swap_write_unplug(sio);
0327         sio = NULL;
0328     }
0329     if (wbc->swap_plug)
0330         *wbc->swap_plug = sio;
0331 
0332     return 0;
0333 }
0334 
0335 int __swap_writepage(struct page *page, struct writeback_control *wbc,
0336              bio_end_io_t end_write_func)
0337 {
0338     struct bio *bio;
0339     int ret;
0340     struct swap_info_struct *sis = page_swap_info(page);
0341 
0342     VM_BUG_ON_PAGE(!PageSwapCache(page), page);
0343     /*
0344      * ->flags can be updated non-atomicially (scan_swap_map_slots),
0345      * but that will never affect SWP_FS_OPS, so the data_race
0346      * is safe.
0347      */
0348     if (data_race(sis->flags & SWP_FS_OPS))
0349         return swap_writepage_fs(page, wbc);
0350 
0351     ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
0352     if (!ret) {
0353         count_swpout_vm_event(page);
0354         return 0;
0355     }
0356 
0357     bio = bio_alloc(sis->bdev, 1,
0358             REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc),
0359             GFP_NOIO);
0360     bio->bi_iter.bi_sector = swap_page_sector(page);
0361     bio->bi_end_io = end_write_func;
0362     bio_add_page(bio, page, thp_size(page), 0);
0363 
0364     bio_associate_blkg_from_page(bio, page);
0365     count_swpout_vm_event(page);
0366     set_page_writeback(page);
0367     unlock_page(page);
0368     submit_bio(bio);
0369 
0370     return 0;
0371 }
0372 
0373 void swap_write_unplug(struct swap_iocb *sio)
0374 {
0375     struct iov_iter from;
0376     struct address_space *mapping = sio->iocb.ki_filp->f_mapping;
0377     int ret;
0378 
0379     iov_iter_bvec(&from, WRITE, sio->bvec, sio->pages, sio->len);
0380     ret = mapping->a_ops->swap_rw(&sio->iocb, &from);
0381     if (ret != -EIOCBQUEUED)
0382         sio_write_complete(&sio->iocb, ret);
0383 }
0384 
0385 static void sio_read_complete(struct kiocb *iocb, long ret)
0386 {
0387     struct swap_iocb *sio = container_of(iocb, struct swap_iocb, iocb);
0388     int p;
0389 
0390     if (ret == sio->len) {
0391         for (p = 0; p < sio->pages; p++) {
0392             struct page *page = sio->bvec[p].bv_page;
0393 
0394             SetPageUptodate(page);
0395             unlock_page(page);
0396         }
0397         count_vm_events(PSWPIN, sio->pages);
0398     } else {
0399         for (p = 0; p < sio->pages; p++) {
0400             struct page *page = sio->bvec[p].bv_page;
0401 
0402             SetPageError(page);
0403             ClearPageUptodate(page);
0404             unlock_page(page);
0405         }
0406         pr_alert_ratelimited("Read-error on swap-device\n");
0407     }
0408     mempool_free(sio, sio_pool);
0409 }
0410 
0411 static void swap_readpage_fs(struct page *page,
0412                  struct swap_iocb **plug)
0413 {
0414     struct swap_info_struct *sis = page_swap_info(page);
0415     struct swap_iocb *sio = NULL;
0416     loff_t pos = page_file_offset(page);
0417 
0418     if (plug)
0419         sio = *plug;
0420     if (sio) {
0421         if (sio->iocb.ki_filp != sis->swap_file ||
0422             sio->iocb.ki_pos + sio->len != pos) {
0423             swap_read_unplug(sio);
0424             sio = NULL;
0425         }
0426     }
0427     if (!sio) {
0428         sio = mempool_alloc(sio_pool, GFP_KERNEL);
0429         init_sync_kiocb(&sio->iocb, sis->swap_file);
0430         sio->iocb.ki_pos = pos;
0431         sio->iocb.ki_complete = sio_read_complete;
0432         sio->pages = 0;
0433         sio->len = 0;
0434     }
0435     sio->bvec[sio->pages].bv_page = page;
0436     sio->bvec[sio->pages].bv_len = thp_size(page);
0437     sio->bvec[sio->pages].bv_offset = 0;
0438     sio->len += thp_size(page);
0439     sio->pages += 1;
0440     if (sio->pages == ARRAY_SIZE(sio->bvec) || !plug) {
0441         swap_read_unplug(sio);
0442         sio = NULL;
0443     }
0444     if (plug)
0445         *plug = sio;
0446 }
0447 
0448 int swap_readpage(struct page *page, bool synchronous,
0449           struct swap_iocb **plug)
0450 {
0451     struct bio *bio;
0452     int ret = 0;
0453     struct swap_info_struct *sis = page_swap_info(page);
0454     bool workingset = PageWorkingset(page);
0455     unsigned long pflags;
0456 
0457     VM_BUG_ON_PAGE(!PageSwapCache(page) && !synchronous, page);
0458     VM_BUG_ON_PAGE(!PageLocked(page), page);
0459     VM_BUG_ON_PAGE(PageUptodate(page), page);
0460 
0461     /*
0462      * Count submission time as memory stall. When the device is congested,
0463      * or the submitting cgroup IO-throttled, submission can be a
0464      * significant part of overall IO time.
0465      */
0466     if (workingset)
0467         psi_memstall_enter(&pflags);
0468     delayacct_swapin_start();
0469 
0470     if (frontswap_load(page) == 0) {
0471         SetPageUptodate(page);
0472         unlock_page(page);
0473         goto out;
0474     }
0475 
0476     if (data_race(sis->flags & SWP_FS_OPS)) {
0477         swap_readpage_fs(page, plug);
0478         goto out;
0479     }
0480 
0481     if (sis->flags & SWP_SYNCHRONOUS_IO) {
0482         ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
0483         if (!ret) {
0484             count_vm_event(PSWPIN);
0485             goto out;
0486         }
0487     }
0488 
0489     ret = 0;
0490     bio = bio_alloc(sis->bdev, 1, REQ_OP_READ, GFP_KERNEL);
0491     bio->bi_iter.bi_sector = swap_page_sector(page);
0492     bio->bi_end_io = end_swap_bio_read;
0493     bio_add_page(bio, page, thp_size(page), 0);
0494     /*
0495      * Keep this task valid during swap readpage because the oom killer may
0496      * attempt to access it in the page fault retry time check.
0497      */
0498     if (synchronous) {
0499         get_task_struct(current);
0500         bio->bi_private = current;
0501     }
0502     count_vm_event(PSWPIN);
0503     bio_get(bio);
0504     submit_bio(bio);
0505     while (synchronous) {
0506         set_current_state(TASK_UNINTERRUPTIBLE);
0507         if (!READ_ONCE(bio->bi_private))
0508             break;
0509 
0510         blk_io_schedule();
0511     }
0512     __set_current_state(TASK_RUNNING);
0513     bio_put(bio);
0514 
0515 out:
0516     if (workingset)
0517         psi_memstall_leave(&pflags);
0518     delayacct_swapin_end();
0519     return ret;
0520 }
0521 
0522 void __swap_read_unplug(struct swap_iocb *sio)
0523 {
0524     struct iov_iter from;
0525     struct address_space *mapping = sio->iocb.ki_filp->f_mapping;
0526     int ret;
0527 
0528     iov_iter_bvec(&from, READ, sio->bvec, sio->pages, sio->len);
0529     ret = mapping->a_ops->swap_rw(&sio->iocb, &from);
0530     if (ret != -EIOCBQUEUED)
0531         sio_read_complete(&sio->iocb, ret);
0532 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!